perf stat: Enable 1ms interval for printing event counters values

From: Alexey Budankov <[email protected]>

Currently print count interval for performance counters values is limited by 10ms so reading the values at frequencies higher than 100Hz is restricted by the tool.

This change makes perf stat -I possible on frequencies up to 1KHz and, to some extent, makes perf stat -I to be on-par with perf record sampling profiling.

When running perf stat -I for monitoring e.g. PCIe uncore counters and at the same time profiling some I/O workload by perf record e.g. for cpu-cycles and context switches, it is then possible to observe consolidated CPU/OS/IO(Uncore) performance picture for that workload.

Tool overhead warning printed when specifying -v option can be missed due to screen scrolling in case you have output to the console so message is moved into help available by running perf stat -h.

---
 tools/perf/Documentation/perf-stat.txt |  2 +-
 tools/perf/builtin-stat.c              | 14 ++------------
 2 files changed, 3 insertions(+), 13 deletions(-)

index f15b306be183..e6c3b4e555c2 100644
@@ -153,7 +153,7 @@ perf stat --repeat 10 --null --sync --pre 'make -s O=defconfig-build/clean' -- m
 
 -I msecs::
 --interval-print msecs::
-Print count deltas every N milliseconds (minimum: 10ms)
+Print count deltas every N milliseconds (minimum: 1ms)
 The overhead percentage could be high in some cases, for instance with small, sub 100ms intervals.  Use with caution.
 	example: 'perf stat -I 1000 -e cycles -a sleep 5'
 

index f5c454855908..147a27e8c937 100644
@@ -1943,7 +1943,8 @@ static const struct option stat_options[] = {
 	OPT_STRING(0, "post", &post_cmd, "command",
 			"command to run after to the measured command"),
 	OPT_UINTEGER('I', "interval-print", &stat_config.interval,
-		    "print counts at regular interval in ms (>= 10)"),
+		    "print counts at regular interval in ms "
+		    "(overhead is possible for values <= 100ms)"),
 	OPT_INTEGER(0, "interval-count", &stat_config.times,
 		    "print counts for fixed number of times"),
 	OPT_UINTEGER(0, "timeout", &stat_config.timeout,

@@ -2923,17 +2924,6 @@ int cmd_stat(int argc, const char **argv)
 		}
 	}
 
-	if (interval && interval < 100) {
-		if (interval < 10) {
-			pr_err("print interval must be >= 10ms\n");
-			parse_options_usage(stat_usage, stat_options, "I", 1);
-			goto out;
-		} else
-			pr_warning("print interval < 100ms. "
-				   "The overhead percentage could be high in some cases. "
-				   "Please proceed with caution.\n");
-	}
-
 	if (stat_config.times && interval)
 		interval_count = true;
 	else if (stat_config.times && !interval) {

2.14.3

From: Mark Rutland <[email protected]>

Our userspace <linux/compiler.h> defines READ_ONCE() in a way that clang doesn't like, as we have an anonymous union in which neither field is initialized.

WRITE_ONCE() is fine since it initializes the __val field. For READ_ONCE() we can keep clang and GCC happy with a dummy initialization of the __c field, so let's do that.

At the same time, let's split READ_ONCE() and WRITE_ONCE() over several lines for legibility, as we do in the in-kernel <linux/compiler.h>.

---
 tools/include/linux/compiler.h | 20 +++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)

index 04e32f965ad7..1827c2f973f9 100644
@@ -151,11 +151,21 @@ static __always_inline void __write_once_size(volatile void *p, void *res, int s
  * required ordering.
  */
 
-#define READ_ONCE(x) \
-	({ union { typeof(x) __val; char __c[1]; } __u; __read_once_size(&(x), __u.__c, sizeof(x)); __u.__val; })
-
-#define WRITE_ONCE(x, val) \
-	({ union { typeof(x) __val; char __c[1]; } __u = { .__val = (val) }; __write_once_size(&(x), __u.__c, sizeof(x)); __u.__val; })
+#define READ_ONCE(x)					\
+({							\
+	union { typeof(x) __val; char __c[1]; } __u =	\
+		{ .__c = { 0 } };			\
+	__read_once_size(&(x), __u.__c, sizeof(x));	\
+	__u.__val;					\
+})
+
+#define WRITE_ONCE(x, val)				\
+({							\
+	union { typeof(x) __val; char __c[1]; } __u =	\
+		{ .__val = (val) }; 			\
+	__write_once_size(&(x), __u.__c, sizeof(x));	\
+	__u.__val;					\
+})
 
 
 #ifndef __fallthrough

2.14.3

From: Kim Phillips <[email protected]>

Enable the unwind test on arm32:

  $ perf test unwind
  58: DWARF unwind                                          : Ok

---
 tools/perf/arch/arm/include/arch-tests.h | 12 ++++++++++++
 tools/perf/arch/arm/tests/Build          |  2 ++
 tools/perf/arch/arm/tests/arch-tests.c   | 16 ++++++++++++++++
 3 files changed, 30 insertions(+)
 create mode 100644 tools/perf/arch/arm/include/arch-tests.h
 create mode 100644 tools/perf/arch/arm/tests/arch-tests.c

new file mode 100644
index 000000000000..90ec4c8cb880
@@ -0,0 +1,12 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef ARCH_TESTS_H
+#define ARCH_TESTS_H
+
+#ifdef HAVE_DWARF_UNWIND_SUPPORT
+struct thread;
+struct perf_sample;
+#endif
+
+extern struct test arch_tests[];
+
+#endif

index b30eff9bcc83..883c57ff0c08 100644
@@ -1,2 +1,4 @@
 libperf-y += regs_load.o
 libperf-y += dwarf-unwind.o
+
+libperf-y += arch-tests.o

new file mode 100644
index 000000000000..5b1543c98022
@@ -0,0 +1,16 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <string.h>
+#include "tests/tests.h"
+#include "arch-tests.h"
+
+struct test arch_tests[] = {
+#ifdef HAVE_DWARF_UNWIND_SUPPORT
+	{
+		.desc = "DWARF unwind",
+		.func = test__dwarf_unwind,
+	},
+#endif
+	{
+		.func = NULL,
+	},
+};

2.14.3

From: Arnaldo Carvalho de Melo <[email protected]>

Jesper wanted to see offsets at callq sites when doing some performance investigation related to retpolines, so save him some time by providing an 'struct annotation_options' to control where offsets should appear:
just on jump targets? That + call instructions? All?

This puts in place the logic to show the offsets, now we need to wire this up in the TUI browser (next patch) and on the 'perf annotate --stdio2" interface, where we need a more general mechanism to setup the 'annotation_options' struct from the command line.

---
 tools/perf/util/annotate.c | 11 +++++++++--
 tools/perf/util/annotate.h |  9 +++++++++
 2 files changed, 18 insertions(+), 2 deletions(-)

index fbad8dfbb186..5edc565d86c4 100644
@@ -46,6 +46,7 @@
 struct annotation_options annotation__default_options = {
 	.use_offset     = true,
 	.jump_arrows    = true,
+	.offset_level	= ANNOTATION__OFFSET_JUMP_TARGETS,
 };
 
 const char 	*disassembler_style;

@@ -2512,7 +2513,8 @@ static void __annotation_line__write(struct annotation_line *al, struct annotati
 		if (!notes->options->use_offset) {
 			printed = scnprintf(bf, sizeof(bf), "%" PRIx64 ": ", addr);
 		} else {
-			if (al->jump_sources) {
+			if (al->jump_sources &&
+			    notes->options->offset_level >= ANNOTATION__OFFSET_JUMP_TARGETS) {
 				if (notes->options->show_nr_jumps) {
 					int prev;
 					printed = scnprintf(bf, sizeof(bf), "%*d ",

@@ -2523,9 +2525,14 @@ static void __annotation_line__write(struct annotation_line *al, struct annotati
 					obj__printf(obj, bf);
 					obj__set_color(obj, prev);
 				}
-
+print_addr:
 				printed = scnprintf(bf, sizeof(bf), "%*" PRIx64 ": ",
 						    notes->widths.target, addr);
+			} else if (ins__is_call(&disasm_line(al)->ins) &&
+				   notes->options->offset_level >= ANNOTATION__OFFSET_CALL) {
+				goto print_addr;
+			} else if (notes->options->offset_level == ANNOTATION__MAX_OFFSET_LEVEL) {
+				goto print_addr;
 			} else {
 				printed = scnprintf(bf, sizeof(bf), "%-*s  ",
 						    notes->widths.addr, " ");

index db8d09bea07e..f28a9e43421d 100644
@@ -70,8 +70,17 @@ struct annotation_options {
 	     show_nr_jumps,
 	     show_nr_samples,
 	     show_total_period;
+	u8   offset_level;
 };
 
+enum {
+	ANNOTATION__OFFSET_JUMP_TARGETS = 1,
+	ANNOTATION__OFFSET_CALL,
+	ANNOTATION__MAX_OFFSET_LEVEL,
+};
+
+#define ANNOTATION__MIN_OFFSET_LEVEL ANNOTATION__OFFSET_JUMP_TARGETS
+
 extern struct annotation_options annotation__default_options;
 
 struct annotation;

2.14.3

From: Arnaldo Carvalho de Melo <[email protected]>

Jesper wanted to see offsets at callq sites when doing some performance investigation related to retpolines, so save him some time by providing a 'O' hotkey to allow showing offsets from function start at call instructions or in all instructions, just go on pressing 'O' till the offsets you need appear.

Example:

Starts with:

Samples: 64 of event 'cycles:ppp', 100000 Hz, Event count (approx.): 318963 ixgbe_read_reg /proc/kcore

Percent│    ↑ je     2a
       │   ┌──cmp    $0xffffffff,%r13d
       │   ├──je     d0
       │   │  mov    $0x53e3,%edi
       │   │→ callq  __const_udelay
       │   │  sub    $0x1,%r15d
       │   │↑ jne    83
       │   │  mov    0x8(%rbp),%rax
       │   │  testb  $0x20,0x1799(%rax)
       │   │↑ je     2a
       │   │  mov    0x200(%rax),%rdi
       │   │  mov    %r13d,%edx
       │   │  mov    $0xffffffffc02595d8,%rsi
       │   │→ callq  netdev_warn
       │   │↑ jmpq   2a
       │d0:└─→mov    0x8(%rbp),%rsi
       │      mov    %rbp,%rdi
       │      mov    %eax,0x4(%rsp)
       │    → callq  ixgbe_remove_adapter.isra.77
       │      mov    0x4(%rsp),%eax

Press 'h' for help on key bindings
============================================================================

Pess 'O':

Samples: 64 of event 'cycles:ppp', 100000 Hz, Event count (approx.): 318963 ixgbe_read_reg /proc/kcore

Percent│    ↑ je     2a
       │   ┌──cmp    $0xffffffff,%r13d
       │   ├──je     d0
       │   │  mov    $0x53e3,%edi
       │99:│→ callq  __const_udelay
       │   │  sub    $0x1,%r15d
       │   │↑ jne    83
       │   │  mov    0x8(%rbp),%rax
       │   │  testb  $0x20,0x1799(%rax)
       │   │↑ je     2a
       │   │  mov    0x200(%rax),%rdi
       │   │  mov    %r13d,%edx
       │   │  mov    $0xffffffffc02595d8,%rsi
       │c6:│→ callq  netdev_warn
       │   │↑ jmpq   2a
       │d0:└─→mov    0x8(%rbp),%rsi
       │      mov    %rbp,%rdi
       │      mov    %eax,0x4(%rsp)
       │db: → callq  ixgbe_remove_adapter.isra.77
       │      mov    0x4(%rsp),%eax

Press 'h' for help on key bindings
============================================================================

Press 'O' again:

Samples: 64 of event 'cycles:ppp', 100000 Hz, Event count (approx.): 318963 ixgbe_read_reg /proc/kcore

Percent│8c: ↑ je     2a
       │8e:┌──cmp    $0xffffffff,%r13d
       │92:├──je     d0
       │94:│  mov    $0x53e3,%edi
       │99:│→ callq  __const_udelay
       │9e:│  sub    $0x1,%r15d
       │a2:│↑ jne    83
       │a4:│  mov    0x8(%rbp),%rax
       │a8:│  testb  $0x20,0x1799(%rax)
       │af:│↑ je     2a
       │b5:│  mov    0x200(%rax),%rdi
       │bc:│  mov    %r13d,%edx
       │bf:│  mov    $0xffffffffc02595d8,%rsi
       │c6:│→ callq  netdev_warn
       │cb:│↑ jmpq   2a
       │d0:└─→mov    0x8(%rbp),%rsi
       │d4:   mov    %rbp,%rdi
       │d7:   mov    %eax,0x4(%rsp)
       │db: → callq  ixgbe_remove_adapter.isra.77
       │e0:   mov    0x4(%rsp),%eax

Press 'h' for help on key bindings
============================================================================

Press 'O' again and it will show just jump target offsets.

---
 tools/perf/ui/browsers/annotate.c | 5 +++++
 1 file changed, 5 insertions(+)

index 12c099a87f8b..3781d74088a7 100644
@@ -692,6 +692,7 @@ static int annotate_browser__run(struct annotate_browser *browser,
 		"J             Toggle showing number of jump sources on targets\n"
 		"n             Search next string\n"
 		"o             Toggle disassembler output/simplified view\n"
+		"O             Bump offset level (jump targets -> +call -> all -> cycle thru)\n"
 		"s             Toggle source code view\n"
 		"t             Circulate percent, total period, samples view\n"
 		"/             Search string\n"

@@ -719,6 +720,10 @@ static int annotate_browser__run(struct annotate_browser *browser,
 			notes->options->use_offset = !notes->options->use_offset;
 			annotation__update_column_widths(notes);
 			continue;
+		case 'O':
+			if (++notes->options->offset_level > ANNOTATION__MAX_OFFSET_LEVEL)
+				notes->options->offset_level = ANNOTATION__MIN_OFFSET_LEVEL;
+			continue;
 		case 'j':
 			notes->options->jump_arrows = !notes->options->jump_arrows;
 			continue;

2.14.3

From: Arnaldo Carvalho de Melo <[email protected]>

Give as examples of package names to install to have this built for fedora and debian, to help the user a bit.

The part from 'e.g.:' onwards:

No openjdk development package found, please install JDK package, e.g. openjdk-8-jdk, java-1.8.0-openjdk-devel

---
 tools/perf/Makefile.config | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

index c7abd83a8e19..6b307e97dc57 100644
@@ -847,7 +847,7 @@ ifndef NO_JVMTI
   ifeq ($(feature-jvmti), 1)
     $(call detected_var,JDIR)
   else
-    $(warning No openjdk development package found, please install JDK package)
+    $(warning No openjdk development package found, please install JDK package, e.g. openjdk-8-jdk, java-1.8.0-openjdk-devel)
     NO_JVMTI := 1
   endif
 endif

2.14.3

From: Arnaldo Carvalho de Melo <[email protected]>

The bpf-script-test-kbuild.c script, used in one of the LLVM subtests, includes ptrace.h unnecessarily, and that ends up making it include a header that uses asm(_ASM_SP), a feature that is not supported by clang <= 4.0, breaking that 'perf test' entry.

This ended up leading to the ca26cffa4e4a ("x86/asm: Allow again using asm.h when building for the 'bpf' clang target"), adding an ifndef __BPF__ to the arch/x86/include/asm/asm.h file.

Newer clang versions accept that asm(_ASM_SP) construct, so just remove the ptrace.h include, which paves the way for reverting ca26cffa4e4a ("x86/asm: Allow again using asm.h when building for the 'bpf' clang target").

---
 tools/perf/tests/bpf-script-test-kbuild.c | 1 -
 1 file changed, 1 deletion(-)

index 3626924740d8..ff3ec8337f0a 100644
@@ -9,7 +9,6 @@
 #define SEC(NAME) __attribute__((section(NAME), used))
 
 #include <uapi/linux/fs.h>
-#include <uapi/asm/ptrace.h>
 
 SEC("func=vfs_llseek")
 int bpf_func__vfs_llseek(void *ctx)

2.14.3

From: Arnaldo Carvalho de Melo <[email protected]>

This reverts commit ca26cffa4e4aaeb09bb9e308f95c7835cb149248.

Newer clang versions accept that asm(_ASM_SP) construct, and now that the bpf-script-test-kbuild.c script, used in one of the 'perf test LLVM' subtests doesn't include ptrace.h, which ended up including arch/x86/include/asm/asm.h, we can revert this patch.

---
 arch/x86/include/asm/asm.h | 2 --
 1 file changed, 2 deletions(-)

index 386a6900e206..219faaec51df 100644
@@ -136,7 +136,6 @@
 #endif
 
 #ifndef __ASSEMBLY__
-#ifndef __BPF__
 /*
  * This output constraint should be used for any inline asm which has a "call"
  * instruction.  Otherwise the asm may be inserted before the frame pointer

@@ -146,6 +145,5 @@
 register unsigned long current_stack_pointer asm(_ASM_SP);
 #define ASM_CALL_CONSTRAINT "+r" (current_stack_pointer)
 #endif
-#endif
 
 #endif /* _ASM_X86_ASM_H */

2.14.3

From: Jin Yao <[email protected]>

To be consistent with other HAVE_XXX_SUPPORT uses in Makefile.config, this patch renames HAVE_SYSCALL_TABLE to HAVE_SYSCALL_TABLE_SUPPORT and updates the C code accordingly.

---
 tools/perf/Makefile.config          | 2 +-
 tools/perf/builtin-help.c           | 2 +-
 tools/perf/perf.c                   | 4 ++--
 tools/perf/util/generate-cmdlist.sh | 2 +-
 tools/perf/util/syscalltbl.c        | 6 +++---
 5 files changed, 8 insertions(+), 8 deletions(-)

index 6b307e97dc57..ae7dc46e8f8a 100644
@@ -68,7 +68,7 @@ ifeq ($(NO_PERF_REGS),0)
 endif
 
 ifneq ($(NO_SYSCALL_TABLE),1)
-  CFLAGS += -DHAVE_SYSCALL_TABLE
+  CFLAGS += -DHAVE_SYSCALL_TABLE_SUPPORT
 endif
 
 # So far there's only x86 and arm libdw unwind support merged in perf.

index 4aca13f23b9d..1c41b4eaf73c 100644
@@ -439,7 +439,7 @@ int cmd_help(int argc, const char **argv)
 #ifdef HAVE_LIBELF_SUPPORT
 		"probe",
 #endif
-#if defined(HAVE_LIBAUDIT_SUPPORT) || defined(HAVE_SYSCALL_TABLE)
+#if defined(HAVE_LIBAUDIT_SUPPORT) || defined(HAVE_SYSCALL_TABLE_SUPPORT)
 		"trace",
 #endif
 	NULL };

index 1659029d03fc..20a08cb32332 100644
@@ -73,7 +73,7 @@ static struct cmd_struct commands[] = {
 	{ "lock",	cmd_lock,	0 },
 	{ "kvm",	cmd_kvm,	0 },
 	{ "test",	cmd_test,	0 },
-#if defined(HAVE_LIBAUDIT_SUPPORT) || defined(HAVE_SYSCALL_TABLE)
+#if defined(HAVE_LIBAUDIT_SUPPORT) || defined(HAVE_SYSCALL_TABLE_SUPPORT)
 	{ "trace",	cmd_trace,	0 },
 #endif
 	{ "inject",	cmd_inject,	0 },

@@ -491,7 +491,7 @@ int main(int argc, const char **argv)
 		argv[0] = cmd;
 	}
 	if (strstarts(cmd, "trace")) {
-#if defined(HAVE_LIBAUDIT_SUPPORT) || defined(HAVE_SYSCALL_TABLE)
+#if defined(HAVE_LIBAUDIT_SUPPORT) || defined(HAVE_SYSCALL_TABLE_SUPPORT)
 		setup_path();
 		argv[0] = "trace";
 		return cmd_trace(argc, argv);

index ff17920a5ebc..c3cef36d4176 100755
@@ -38,7 +38,7 @@ do
 done
 echo "#endif /* HAVE_LIBELF_SUPPORT */"
 
-echo "#if defined(HAVE_LIBAUDIT_SUPPORT) || defined(HAVE_SYSCALL_TABLE)"
+echo "#if defined(HAVE_LIBAUDIT_SUPPORT) || defined(HAVE_SYSCALL_TABLE_SUPPORT)"
 sed -n -e 's/^perf-\([^ 	]*\)[ 	].* audit*/\1/p' command-list.txt |
 sort |
 while read cmd

index 895122d638dd..0ee7f568d60c 100644
@@ -17,7 +17,7 @@
 #include <stdlib.h>
 #include <linux/compiler.h>
 
-#ifdef HAVE_SYSCALL_TABLE
+#ifdef HAVE_SYSCALL_TABLE_SUPPORT
 #include <string.h>
 #include "string2.h"
 #include "util.h"

@@ -139,7 +139,7 @@ int syscalltbl__strglobmatch_first(struct syscalltbl *tbl, const char *syscall_g
 	return syscalltbl__strglobmatch_next(tbl, syscall_glob, idx);
 }
 
-#else /* HAVE_SYSCALL_TABLE */
+#else /* HAVE_SYSCALL_TABLE_SUPPORT */
 
 #include <libaudit.h>
 

@@ -176,4 +176,4 @@ int syscalltbl__strglobmatch_first(struct syscalltbl *tbl, const char *syscall_g
 {
 	return syscalltbl__strglobmatch_next(tbl, syscall_glob, idx);
 }
-#endif /* HAVE_SYSCALL_TABLE */
+#endif /* HAVE_SYSCALL_TABLE_SUPPORT */

2.14.3

From: Jin Yao <[email protected]>

This patch doesn't print "libaudit" line if HAVE_SYSCALL_TABLE_SUPPORT is available and add a line for HAVE_SYSCALL_TABLE_SUPPORT.

For example,

$ ./perf -vv
perf version 4.13.rc5.gc2f8af9
                 dwarf: [ on  ]  # HAVE_DWARF_SUPPORT
    dwarf_getlocations: [ on  ]  # HAVE_DWARF_GETLOCATIONS_SUPPORT
                 glibc: [ on  ]  # HAVE_GLIBC_SUPPORT
                  gtk2: [ on  ]  # HAVE_GTK2_SUPPORT
         syscall_table: [ on  ]  # HAVE_SYSCALL_TABLE_SUPPORT
                libbfd: [ on  ]  # HAVE_LIBBFD_SUPPORT
                libelf: [ on  ]  # HAVE_LIBELF_SUPPORT
               libnuma: [ on  ]  # HAVE_LIBNUMA_SUPPORT
numa_num_possible_cpus: [ on  ]  # HAVE_LIBNUMA_SUPPORT
               libperl: [ on  ]  # HAVE_LIBPERL_SUPPORT
             libpython: [ on  ]  # HAVE_LIBPYTHON_SUPPORT
              libslang: [ on  ]  # HAVE_SLANG_SUPPORT
             libcrypto: [ on  ]  # HAVE_LIBCRYPTO_SUPPORT
             libunwind: [ on  ]  # HAVE_LIBUNWIND_SUPPORT
    libdw-dwarf-unwind: [ on  ]  # HAVE_DWARF_SUPPORT
                  zlib: [ on  ]  # HAVE_ZLIB_SUPPORT
                  lzma: [ on  ]  # HAVE_LZMA_SUPPORT
             get_cpuid: [ on  ]  # HAVE_AUXTRACE_SUPPORT
                   bpf: [ on  ]  # HAVE_LIBBPF_SUPPORT

The line "syscall_table: [ on ] # HAVE_SYSCALL_TABLE_SUPPORT" is new created.

---
 tools/perf/builtin-version.c | 3 +++
 1 file changed, 3 insertions(+)

index 2abe3910d6b6..50df168be326 100644
@@ -60,7 +60,10 @@ static void library_status(void)
 	STATUS(HAVE_DWARF_GETLOCATIONS_SUPPORT, dwarf_getlocations);
 	STATUS(HAVE_GLIBC_SUPPORT, glibc);
 	STATUS(HAVE_GTK2_SUPPORT, gtk2);
+#ifndef HAVE_SYSCALL_TABLE_SUPPORT
 	STATUS(HAVE_LIBAUDIT_SUPPORT, libaudit);
+#endif
+	STATUS(HAVE_SYSCALL_TABLE_SUPPORT, syscall_table);
 	STATUS(HAVE_LIBBFD_SUPPORT, libbfd);
 	STATUS(HAVE_LIBELF_SUPPORT, libelf);
 	STATUS(HAVE_LIBNUMA_SUPPORT, libnuma);

2.14.3

From: Takuya Yamamoto <[email protected]>

Fixed a incorrect option and usage to those shown by "perf sched timehist -h", i.e. the default is really --call-graph, which is equivalent to -g.

---
 tools/perf/Documentation/perf-sched.txt | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

index bb33601a823b..63f938b887dd 100644
@@ -104,8 +104,8 @@ OPTIONS for 'perf sched timehist'
     kallsyms pathname
 
 -g::
---no-call-graph::
-	Do not display call chains if present.
+--call-graph::
+	Display call chains if present (default on).
 
 --max-stack::
 	Maximum number of functions to display in backtrace, default 5.

2.14.3

From: Sandipan Das <[email protected]>

We disable this test as instruction breakpoints (HW_BREAKPOINT_X) are not available for powerpc.

Before applying patch:

21: Breakpoint accounting :
--- start ---
test child forked, pid 3635
failed opening event 0
failed opening event 0
watchpoints count 1, breakpoints count 0, has_ioctl 1, share 0 test child finished with -2
---- end ----
Breakpoint accounting: Skip

After applying patch:

21: Breakpoint accounting : Disabled

---
 tools/perf/tests/builtin-test.c | 1 +
 1 file changed, 1 insertion(+)

index 625f5a6772af..cac8f8889bc3 100644
@@ -118,6 +118,7 @@ static struct test generic_tests[] = {
 	{
 		.desc = "Breakpoint accounting",
 		.func = test__bp_accounting,
+		.is_supported = test__bp_signal_is_supported,
 	},
 	{
 		.desc = "Number of exit events of a simple workload",

2.14.3

From: Thomas Richter <[email protected]>

Using perf on 4.16.0 kernel on s390 shows this warning:

failed: can't open node sysfs data

each time I run command perf record ... for example:

[root@s35lp76 perf]# ./perf record -e rB0000 -- sleep 1 [ perf record: Woken up 1 times to write data ]
failed: can't open node sysfs data
[ perf record: Captured and wrote 0.001 MB perf.data (4 samples) ] [root@s35lp76 perf]#

It turns out commit e2091cedd51bf ("perf tools: Add MEM_TOPOLOGY feature to perf data file") tries to open directory named /sys/devices/system/node/ which does not exist on s390.

This is the call stack:
__cmd_record

 +---> perf_session__write_header
       +---> perf_header__adds_write
             +---> do_write_feat
	           +---> write_mem_topology
		         +---> build_mem_topology

prints warning

The issue starts in do_write_feat() which unconditionally loops over all features and now includes HEADER_MEM_TOPOLOGY and calls write_mem_topology().

Function record__init_features() at the beginning of __cmd_record() sets all features and then turns off some of them.

Fix this by changing the warning to a level 2 debug output statement.

So it is only shown when debug level 2 or higher is set.

---
 tools/perf/util/header.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

index 121df1683c36..a8bff2178fbc 100644
@@ -1320,7 +1320,8 @@ static int build_mem_topology(struct memory_node *nodes, u64 size, u64 *cntp)
 
 	dir = opendir(path);
 	if (!dir) {
-		pr_warning("failed: can't open node sysfs data\n");
+		pr_debug2("%s: could't read %s, does this arch have topology information?\n",
+			  __func__, path);
 		return -1;
 	}
 

2.14.3

From: Arnaldo Carvalho de Melo <[email protected]>

In the TUI the 's' hotkey can be used to switch to another perf.data file in the current directory, but that got broken in Fixes:
b01141f4f59c ("perf annotate: Initialize the priv are in symbol__new()"), that would show this once another file was chosen:

    ┌─Fatal Error─────────────────────────────────────┐
    │Annotation needs to be init before symbol__init()│
    │                                                 │
    │                                                 │
    │Press any key...                                 │
    └─────────────────────────────────────────────────┘

Fix it by just silently bailing out if symbol__annotation_init() was already called, just like is done with symbol__init(), i.e. they are done just once at session start, not when switching to a new perf.data file.

---
 tools/perf/util/symbol.c | 8 +++-----
 1 file changed, 3 insertions(+), 5 deletions(-)

index 62b2dd2253eb..1466814ebada 100644
@@ -2091,16 +2091,14 @@ static bool symbol__read_kptr_restrict(void)
 
 int symbol__annotation_init(void)
 {
+	if (symbol_conf.init_annotation)
+		return 0;
+
 	if (symbol_conf.initialized) {
 		pr_err("Annotation needs to be init before symbol__init()\n");
 		return -1;
 	}
 
-	if (symbol_conf.init_annotation) {
-		pr_warning("Annotation being initialized multiple times\n");
-		return 0;
-	}
-
 	symbol_conf.priv_size += sizeof(struct annotation);
 	symbol_conf.init_annotation = true;
 	return 0;

2.14.3

From: Arnaldo Carvalho de Melo <[email protected]>

The default is 1 (jump_target):

  # perf annotate --ignore-vmlinux --stdio2 _raw_spin_lock_irqsave
  Samples: 3K of event 'cycles:ppp', 3000 Hz, Event count (approx.): 2766398574
  _raw_spin_lock_irqsave() /proc/kcore
    0.26        nop
    4.61        push   %rbx
   19.33        pushfq
    7.97        pop    %rax
    0.32        nop
    0.06        mov    %rax,%rbx
   14.63        cli
    0.06        nop
                xor    %eax,%eax
                mov    $0x1,%edx
   49.94        lock   cmpxchg %edx,(%rdi)
    0.16        test   %eax,%eax
              ↓ jne    2b
    2.66        mov    %rbx,%rax
                pop    %rbx
              ← retq
          2b:   mov    %eax,%esi
              → callq  *ffffffffb30eaed0
                mov    %rbx,%rax
                pop    %rbx
              ← retq
  #

But one can ask for showing offsets for call instructions by setting this:

  # perf annotate --ignore-vmlinux --stdio2 _raw_spin_lock_irqsave
  Samples: 3K of event 'cycles:ppp', 3000 Hz, Event count (approx.): 2766398574
  _raw_spin_lock_irqsave() /proc/kcore
    0.26        nop
    4.61        push   %rbx
   19.33        pushfq
    7.97        pop    %rax
    0.32        nop
    0.06        mov    %rax,%rbx
   14.63        cli
    0.06        nop
                xor    %eax,%eax
                mov    $0x1,%edx
   49.94        lock   cmpxchg %edx,(%rdi)
    0.16        test   %eax,%eax
              ↓ jne    2b
    2.66        mov    %rbx,%rax
                pop    %rbx
              ← retq
          2b:   mov    %eax,%esi
          2d: → callq  *ffffffffb30eaed0
                mov    %rbx,%rax
                pop    %rbx
              ← retq
  #

Or using a big value to ask for all offsets to be shown:

  # cat ~/.perfconfig
  [annotate]

	offset_level = 100

	hide_src_code = true
  # perf annotate --ignore-vmlinux --stdio2 _raw_spin_lock_irqsave
  Samples: 3K of event 'cycles:ppp', 3000 Hz, Event count (approx.): 2766398574
  _raw_spin_lock_irqsave() /proc/kcore
    0.26   0:   nop
    4.61   5:   push   %rbx
   19.33   6:   pushfq
    7.97   7:   pop    %rax
    0.32   8:   nop
    0.06   d:   mov    %rax,%rbx
   14.63  10:   cli
    0.06  11:   nop
          17:   xor    %eax,%eax
          19:   mov    $0x1,%edx
   49.94  1e:   lock   cmpxchg %edx,(%rdi)
    0.16  22:   test   %eax,%eax
          24: ↓ jne    2b
    2.66  26:   mov    %rbx,%rax
          29:   pop    %rbx
          2a: ← retq
          2b:   mov    %eax,%esi
          2d: → callq  *ffffffffb30eaed0
          32:   mov    %rbx,%rax
          35:   pop    %rbx
          36: ← retq
   #

This also affects the TUI, i.e. the default 'perf annotate' and 'perf top/report' -> A hotkey -> annotate interfaces, when slang-devel is present in the build, i.e.:

  # perf version --build-options | grep slang
              libslang: [ on  ]  # HAVE_SLANG_SUPPORT
  #

---
 tools/perf/Documentation/perf-config.txt |  5 +++++
 tools/perf/util/annotate.c               | 15 ++++++++++++---
 2 files changed, 17 insertions(+), 3 deletions(-)

index 5b4fff3adc4b..32f4a898e3f2 100644
@@ -334,6 +334,11 @@ annotate.*::
 
 		99.93 │      mov    %eax,%eax
 
+	annotate.offset_level::
+		Default is '1', meaning just jump targets will have offsets show right beside
+		the instruction. When set to '2' 'call' instructions will also have its offsets
+		shown, 3 or higher will show offsets for all instructions.
+
 hist.*::
 	hist.percentage::
 		This option control the way to calculate overhead of filtered entries -

index 5edc565d86c4..536ee148bff8 100644
@@ -2649,10 +2649,11 @@ int __annotation__scnprintf_samples_period(struct annotation *notes,
  */
 static struct annotation_config {
 	const char *name;
-	bool *value;
+	void *value;
 } annotation__configs[] = {
 	ANNOTATION__CFG(hide_src_code),
 	ANNOTATION__CFG(jump_arrows),
+	ANNOTATION__CFG(offset_level),
 	ANNOTATION__CFG(show_linenr),
 	ANNOTATION__CFG(show_nr_jumps),
 	ANNOTATION__CFG(show_nr_samples),

@@ -2684,8 +2685,16 @@ static int annotation__config(const char *var, const char *value,
 
 	if (cfg == NULL)
 		pr_debug("%s variable unknown, ignoring...", var);
-	else
-		*cfg->value = perf_config_bool(name, value);
+	else if (strcmp(var, "annotate.offset_level") == 0) {
+		perf_config_int(cfg->value, name, value);
+
+		if (*(int *)cfg->value > ANNOTATION__MAX_OFFSET_LEVEL)
+			*(int *)cfg->value = ANNOTATION__MAX_OFFSET_LEVEL;
+		else if (*(int *)cfg->value < ANNOTATION__MIN_OFFSET_LEVEL)
+			*(int *)cfg->value = ANNOTATION__MIN_OFFSET_LEVEL;
+	} else {
+		*(bool *)cfg->value = perf_config_bool(name, value);
+	}
 	return 0;
 }
 

2.14.3

From: Arnaldo Carvalho de Melo <[email protected]>

Just like is done for 'mov' and others that can have as source or targets variables resolved by objdump, to make them more compact:

-               orb    $0x4,0x224d71(%rip)        # 226ca4 <_rtld_global+0xca4>
+               orb    $0x4,_rtld_global+0xca4

---
 tools/perf/arch/x86/annotate/instructions.c | 67 ++++++++++++++++++++++++++++-
 1 file changed, 66 insertions(+), 1 deletion(-)

index 5bd1ba8c0282..44f5aba78210 100644
@@ -1,21 +1,43 @@
 // SPDX-License-Identifier: GPL-2.0
 static struct ins x86__instructions[] = {
+	{ .name = "adc",	.ops = &mov_ops,  },
+	{ .name = "adcb",	.ops = &mov_ops,  },
+	{ .name = "adcl",	.ops = &mov_ops,  },
 	{ .name = "add",	.ops = &mov_ops,  },
 	{ .name = "addl",	.ops = &mov_ops,  },
 	{ .name = "addq",	.ops = &mov_ops,  },
+	{ .name = "addsd",	.ops = &mov_ops,  },
 	{ .name = "addw",	.ops = &mov_ops,  },
 	{ .name = "and",	.ops = &mov_ops,  },
+	{ .name = "andb",	.ops = &mov_ops,  },
+	{ .name = "andl",	.ops = &mov_ops,  },
+	{ .name = "andpd",	.ops = &mov_ops,  },
+	{ .name = "andps",	.ops = &mov_ops,  },
+	{ .name = "andq",	.ops = &mov_ops,  },
+	{ .name = "andw",	.ops = &mov_ops,  },
+	{ .name = "bsr",	.ops = &mov_ops,  },
+	{ .name = "bt",		.ops = &mov_ops,  },
+	{ .name = "btr",	.ops = &mov_ops,  },
 	{ .name = "bts",	.ops = &mov_ops,  },
+	{ .name = "btsq",	.ops = &mov_ops,  },
 	{ .name = "call",	.ops = &call_ops, },
 	{ .name = "callq",	.ops = &call_ops, },
+	{ .name = "cmovbe",	.ops = &mov_ops,  },
+	{ .name = "cmove",	.ops = &mov_ops,  },
+	{ .name = "cmovae",	.ops = &mov_ops,  },
 	{ .name = "cmp",	.ops = &mov_ops,  },
 	{ .name = "cmpb",	.ops = &mov_ops,  },
 	{ .name = "cmpl",	.ops = &mov_ops,  },
 	{ .name = "cmpq",	.ops = &mov_ops,  },
 	{ .name = "cmpw",	.ops = &mov_ops,  },
 	{ .name = "cmpxch",	.ops = &mov_ops,  },
+	{ .name = "cmpxchg",	.ops = &mov_ops,  },
+	{ .name = "cs",		.ops = &mov_ops,  },
 	{ .name = "dec",	.ops = &dec_ops,  },
 	{ .name = "decl",	.ops = &dec_ops,  },
+	{ .name = "divsd",	.ops = &mov_ops,  },
+	{ .name = "divss",	.ops = &mov_ops,  },
+	{ .name = "gs",		.ops = &mov_ops,  },
 	{ .name = "imul",	.ops = &mov_ops,  },
 	{ .name = "inc",	.ops = &dec_ops,  },
 	{ .name = "incl",	.ops = &dec_ops,  },

@@ -57,25 +79,68 @@ static struct ins x86__instructions[] = {
 	{ .name = "lea",	.ops = &mov_ops,  },
 	{ .name = "lock",	.ops = &lock_ops, },
 	{ .name = "mov",	.ops = &mov_ops,  },
+	{ .name = "movapd",	.ops = &mov_ops,  },
+	{ .name = "movaps",	.ops = &mov_ops,  },
 	{ .name = "movb",	.ops = &mov_ops,  },
 	{ .name = "movdqa",	.ops = &mov_ops,  },
+	{ .name = "movdqu",	.ops = &mov_ops,  },
 	{ .name = "movl",	.ops = &mov_ops,  },
 	{ .name = "movq",	.ops = &mov_ops,  },
+	{ .name = "movsd",	.ops = &mov_ops,  },
 	{ .name = "movslq",	.ops = &mov_ops,  },
+	{ .name = "movss",	.ops = &mov_ops,  },
+	{ .name = "movupd",	.ops = &mov_ops,  },
+	{ .name = "movups",	.ops = &mov_ops,  },
+	{ .name = "movw",	.ops = &mov_ops,  },
 	{ .name = "movzbl",	.ops = &mov_ops,  },
 	{ .name = "movzwl",	.ops = &mov_ops,  },
+	{ .name = "mulsd",	.ops = &mov_ops,  },
+	{ .name = "mulss",	.ops = &mov_ops,  },
 	{ .name = "nop",	.ops = &nop_ops,  },
 	{ .name = "nopl",	.ops = &nop_ops,  },
 	{ .name = "nopw",	.ops = &nop_ops,  },
 	{ .name = "or",		.ops = &mov_ops,  },
+	{ .name = "orb",	.ops = &mov_ops,  },
 	{ .name = "orl",	.ops = &mov_ops,  },
+	{ .name = "orps",	.ops = &mov_ops,  },
+	{ .name = "orq",	.ops = &mov_ops,  },
+	{ .name = "pand",	.ops = &mov_ops,  },
+	{ .name = "paddq",	.ops = &mov_ops,  },
+	{ .name = "pcmpeqb",	.ops = &mov_ops,  },
+	{ .name = "por",	.ops = &mov_ops,  },
+	{ .name = "rclb",	.ops = &mov_ops,  },
+	{ .name = "rcll",	.ops = &mov_ops,  },
+	{ .name = "retq",	.ops = &ret_ops,  },
+	{ .name = "sbb",	.ops = &mov_ops,  },
+	{ .name = "sbbl",	.ops = &mov_ops,  },
+	{ .name = "sete",	.ops = &mov_ops,  },
+	{ .name = "sub",	.ops = &mov_ops,  },
+	{ .name = "subl",	.ops = &mov_ops,  },
+	{ .name = "subq",	.ops = &mov_ops,  },
+	{ .name = "subsd",	.ops = &mov_ops,  },
+	{ .name = "subw",	.ops = &mov_ops,  },
 	{ .name = "test",	.ops = &mov_ops,  },
 	{ .name = "testb",	.ops = &mov_ops,  },
 	{ .name = "testl",	.ops = &mov_ops,  },
+	{ .name = "ucomisd",	.ops = &mov_ops,  },
+	{ .name = "ucomiss",	.ops = &mov_ops,  },
+	{ .name = "vaddsd",	.ops = &mov_ops,  },
+	{ .name = "vandpd",	.ops = &mov_ops,  },
+	{ .name = "vmovdqa",	.ops = &mov_ops,  },
+	{ .name = "vmovq",	.ops = &mov_ops,  },
+	{ .name = "vmovsd",	.ops = &mov_ops,  },
+	{ .name = "vmulsd",	.ops = &mov_ops,  },
+	{ .name = "vorpd",	.ops = &mov_ops,  },
+	{ .name = "vsubsd",	.ops = &mov_ops,  },
+	{ .name = "vucomisd",	.ops = &mov_ops,  },
 	{ .name = "xadd",	.ops = &mov_ops,  },
 	{ .name = "xbeginl",	.ops = &jump_ops, },
 	{ .name = "xbeginq",	.ops = &jump_ops, },
-	{ .name = "retq",	.ops = &ret_ops,  },
+	{ .name = "xchg",	.ops = &mov_ops,  },
+	{ .name = "xor",	.ops = &mov_ops, },
+	{ .name = "xorb",	.ops = &mov_ops, },
+	{ .name = "xorpd",	.ops = &mov_ops, },
+	{ .name = "xorps",	.ops = &mov_ops, },
 };
 
 static bool x86__ins_is_fused(struct arch *arch, const char *ins1,

2.14.3

Em Fri, Apr 13, 2018 at 09:20:43AM -0700, Andi Kleen escreveu:

So, the function is _dl_start in /lib64/ld-2.26.so, the objdump output is:

0000000000001b10 <_dl_start>:
<SNIP>

    1d1f:       0f 84 ab 00 00 00       je     1dd0 <_dl_start+0x2c0>
    1d25:       48 8d 3d 64 3c 22 00    lea    0x223c64(%rip),%rdi        # 225990 <_rtld_global+0x990>
    1d2c:       80 0d 71 3f 22 00 04    orb    $0x4,0x223f71(%rip)        # 225ca4 <_rtld_global+0xca4>
    1d33:       e8 78 92 00 00          callq  afb0 <_dl_setup_hash>

<SNIP>

Then...

    21: 0000000000225000  3960 OBJECT  GLOBAL DEFAULT   20 _rtld_global@@GLIBC_PRIVATE

0x225ca4 = 0x225000 + 0x0xca4

And that is equal to 0x1d33 + 0x223f71

What do I miss? Or where is it that I'm misinterpreting the calculations that objdump did in its output?

This is just to make things compact, on the TUI if one wants to see the original its just a matter of pressing 'o', i.e. that line appears by default as:

perf annotate _dl_start
press HOME
press /orb
<SNIP>

       │215:   lea    _rtld_global+0x990,%rdi
       │       orb    $0x4,_rtld_global+0xca4
       │     → callq  _dl_setup_hash 

<SNIP>

press 'o', to see details, i.e. was this encoded as rip-relative?

<SNIP>

       │1f25:   lea    0x224a64(%rip),%rdi        # 226990 <_rtld_global+0x990>
       │1f2c:   orb    $0x4,0x224d71(%rip)        # 226ca4 <_rtld_global+0xca4>
       │1f33: → callq  b080 <_dl_setup_hash>  

<SNIP>

Now, if we have a non-rip relative orb, lemme search one with:

perf annotate --stdio2
/orb

There, we have one in the kernel's sys_pselect6:

perf annotate sys_pselect6
/orb

       │       mov    %gs:0x15b40,%rax
       │       orb    $0x4,0x490(%rax)
       │       mov    %rdx,0x718(%rax)
       │       mov    (%rax),%rdx

No variables resolved, nothing done, press 'o' to see the original objdump disassembled line:

       │ffffffff8128af29:   mov    %gs:0x15b40,%rax
       │ffffffff8128af32:   orb    $0x4,0x490(%rax)
       │ffffffff8128af39:   mov    %rdx,0x718(%rax)
       │ffffffff8128af40:   mov    (%rax),%rdx

About something mildly related: what do you think about this:
http://ref.x86asm.net/, there is a xml file there[1] I'm thinking about using, if available on the developer's HOME or some other standard place, to provide help about the instructions :-)

- Arnaldo

[1] http://ref.x86asm.net/x86reference.xml

look for <mnem>PUNPCKHQDQ</mnem>

<entry r="yes">

	<pref>66</pref>
	<proc_start>10</proc_start>
	<syntax>
		<mnem>PUNPCKHQDQ</mnem>
		<dst>
			<a>V</a>
			<t>dq</t>
		</dst>
		<src>
			<a>W</a>
			<t>dq</t>
		</src>
	</syntax>
	<instr_ext>sse2</instr_ext>
	<grp1>simdint</grp1>
	<grp2>shunpck</grp2>
	<note>
		<brief>Unpack High Data</brief>
	</note>

</entry>

8-)

The calculations are right, but these are still two different address modes. You cannot just turn one silently into the other.

I think it would be ok to use the syntax in the assembler

symbol(%rip) with no # ...

I don't know how well it's going to be maintained. x86 changes a lot and I've seen a lot of disassembler libraries etc. go stale as the owner cannot keep up.

The only semi official maintained descriptions are the XED tables (but those don't have descriptions) or the PDFs from Intel/AMD.
I suppose could have some hack that talks to a PDF reader and automatically downloads/searches the PDF.

If unofficial is ok I would rather port some functionality from https://github.com/HJLebbink/asm-dude
which has a lot of cool stuff.

-Andi

Hi,

One thing I find useful is to show string constant if the address is in the rodata (and printable of course). Maybe something like below..

lea 0x1234(%rip),%rdi # "hello world\n"
callq printf

Just an idea.

Thanks,
Namhyung

Em Tue, Apr 17, 2018 at 11:56:28AM +0900, Namhyung Kim escreveu:

Send some more :-)

/me adds to the todo.annotate folder...

- Arnaldo