Skip to contentSkip to conversation

csky: perf callchain dwarf support

PATCH Superseded Mao Han 3 participants 10 messages Apr 10, 2019 – Apr 16, 2019

Subsystems (2)

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore

Mao Han

author
#0 lore raw

v3 cover csky: perf callchain dwarf support

This patch set add perf DWARF unwinding support for C-SKY. Including user registers/stack dump API, and libdw support.

CC Peter Zijlstra <[email protected]>
CC Ingo Molnar <[email protected]>
CC Arnaldo Carvalho de Melo <[email protected]>
CC Alexander Shishkin <[email protected]>
CC Jiri Olsa <[email protected]>
CC Namhyung Kim <[email protected]>
CC Guo Ren <[email protected]>

Changes since v2:
- use same registers name as struct pt_regs
- code convention

Changes since v1:
- seperate the callchain support using frame pointer

Mao Han (3):
perf: use hweight64 instead of hweight_long
csky: Add support for perf registers sampling
csky: add support for libdw

 arch/csky/Kconfig                            |   2 +
 arch/csky/include/uapi/asm/perf_regs.h       |  51 ++++++++++++++
 arch/csky/kernel/Makefile                    |   1 +
 arch/csky/kernel/perf_regs.c                 |  40 +++++++++++
 tools/arch/csky/include/uapi/asm/perf_regs.h |  51 ++++++++++++++
 tools/perf/Makefile.config                   |   6 +-
 tools/perf/arch/csky/Build                   |   1 +
 tools/perf/arch/csky/Makefile                |   3 +
 tools/perf/arch/csky/include/perf_regs.h     | 100 +++++++++++++++++++++++++++
 tools/perf/arch/csky/util/Build              |   2 +
 tools/perf/arch/csky/util/dwarf-regs.c       |  49 +++++++++++++
 tools/perf/arch/csky/util/unwind-libdw.c     |  78 +++++++++++++++++++++
 tools/perf/util/evsel.c                      |   2 +-
 13 files changed, 384 insertions(+), 2 deletions(-)
 create mode 100644 arch/csky/include/uapi/asm/perf_regs.h
 create mode 100644 arch/csky/kernel/perf_regs.c
 create mode 100644 tools/arch/csky/include/uapi/asm/perf_regs.h
 create mode 100644 tools/perf/arch/csky/Build
 create mode 100644 tools/perf/arch/csky/Makefile
 create mode 100644 tools/perf/arch/csky/include/perf_regs.h
 create mode 100644 tools/perf/arch/csky/util/Build
 create mode 100644 tools/perf/arch/csky/util/dwarf-regs.c
 create mode 100644 tools/perf/arch/csky/util/unwind-libdw.c
Signature · 1 line
2.7.4
author
#8 lore raw

PATCH v3 2/3 csky: Add support for perf registers sampling

This patch implements the perf registers sampling and validation API for csky arch. The valid registers and their register ID are defined in perf_regs.h. Perf tool can backtrace in userspace with unwind library and the registers/user stack dump support.

CC Guo Ren <[email protected]>
---
 arch/csky/Kconfig                      |  2 ++
 arch/csky/include/uapi/asm/perf_regs.h | 51 ++++++++++++++++++++++++++++++++++
 arch/csky/kernel/Makefile              |  1 +
 arch/csky/kernel/perf_regs.c           | 40 ++++++++++++++++++++++++++
 4 files changed, 94 insertions(+)
 create mode 100644 arch/csky/include/uapi/asm/perf_regs.h
 create mode 100644 arch/csky/kernel/perf_regs.c
arch/csky/Kconfig +2 -0
index c4974cf..8e45c7a 100644
@@ -38,6 +38,8 @@ config CSKY
 	select HAVE_KERNEL_LZO
 	select HAVE_KERNEL_LZMA
 	select HAVE_PERF_EVENTS
+	select HAVE_PERF_REGS
+	select HAVE_PERF_USER_STACK_DUMP
 	select HAVE_DMA_API_DEBUG
 	select HAVE_DMA_CONTIGUOUS
 	select HAVE_SYSCALL_TRACEPOINTS
arch/csky/include/uapi/asm/perf_regs.h new file +51 -0

51 lines not on this page · raw message

arch/csky/kernel/Makefile +1 -0
index 4c462f5..1624b04 100644
@@ -10,6 +10,7 @@ obj-$(CONFIG_FUNCTION_TRACER)		+= ftrace.o
 obj-$(CONFIG_STACKTRACE)		+= stacktrace.o
 obj-$(CONFIG_CSKY_PMU_V1)		+= perf_event.o
 obj-$(CONFIG_PERF_EVENTS)		+= perf_callchain.o
+obj-$(CONFIG_HAVE_PERF_REGS)            += perf_regs.o
 
 ifdef CONFIG_FUNCTION_TRACER
 CFLAGS_REMOVE_ftrace.o = $(CC_FLAGS_FTRACE)
arch/csky/kernel/perf_regs.c new file +40 -0
new file mode 100644
index 0000000..88f1875
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: GPL-2.0
+// Copyright (C) 2019 Hangzhou C-SKY Microsystems co.,ltd.
+
+#include <linux/errno.h>
+#include <linux/kernel.h>
+#include <linux/perf_event.h>
+#include <linux/bug.h>
+#include <asm/perf_regs.h>
+#include <asm/ptrace.h>
+
+u64 perf_reg_value(struct pt_regs *regs, int idx)
+{
+	if (WARN_ON_ONCE((u32)idx >= PERF_REG_CSKY_MAX))
+		return 0;
+
+	return ((long *)regs)[idx];
+}
+
+#define REG_RESERVED (~((1ULL << PERF_REG_CSKY_MAX) - 1))
+
+int perf_reg_validate(u64 mask)
+{
+	if (!mask || mask & REG_RESERVED)
+		return -EINVAL;
+
+	return 0;
+}
+
+u64 perf_reg_abi(struct task_struct *task)
+{
+	return PERF_SAMPLE_REGS_ABI_32;
+}
+
+void perf_get_regs_user(struct perf_regs *regs_user,
+			struct pt_regs *regs,
+			struct pt_regs *regs_user_copy)
+{
+	regs_user->regs = task_pt_regs(current);
+	regs_user->abi = perf_reg_abi(current);
+}
Signature · 1 line
2.7.4
author
#9 lore raw

PATCH v3 3/3 csky: add support for libdw

In reply to Mao Han

This patch add support for DWARF register mappings and libdw registers initialization, which is used by perf callchain analyzing when --call-graph=dwarf is given.

CC Peter Zijlstra <[email protected]>
CC Ingo Molnar <[email protected]>
CC Arnaldo Carvalho de Melo <[email protected]>
CC Alexander Shishkin <[email protected]>
CC Jiri Olsa <[email protected]>
CC Namhyung Kim <[email protected]>
---
 tools/arch/csky/include/uapi/asm/perf_regs.h |  51 ++++++++++++++
 tools/perf/Makefile.config                   |   6 +-
 tools/perf/arch/csky/Build                   |   1 +
 tools/perf/arch/csky/Makefile                |   3 +
 tools/perf/arch/csky/include/perf_regs.h     | 100 +++++++++++++++++++++++++++
 tools/perf/arch/csky/util/Build              |   2 +
 tools/perf/arch/csky/util/dwarf-regs.c       |  49 +++++++++++++
 tools/perf/arch/csky/util/unwind-libdw.c     |  78 +++++++++++++++++++++
 8 files changed, 289 insertions(+), 1 deletion(-)
 create mode 100644 tools/arch/csky/include/uapi/asm/perf_regs.h
 create mode 100644 tools/perf/arch/csky/Build
 create mode 100644 tools/perf/arch/csky/Makefile
 create mode 100644 tools/perf/arch/csky/include/perf_regs.h
 create mode 100644 tools/perf/arch/csky/util/Build
 create mode 100644 tools/perf/arch/csky/util/dwarf-regs.c
 create mode 100644 tools/perf/arch/csky/util/unwind-libdw.c
tools/arch/csky/include/uapi/asm/perf_regs.h new file +51 -0
new file mode 100644
index 0000000..ee323d8
@@ -0,0 +1,51 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+// Copyright (C) 2019 Hangzhou C-SKY Microsystems co.,ltd.
+
+#ifndef _ASM_CSKY_PERF_REGS_H
+#define _ASM_CSKY_PERF_REGS_H
+
+/* Index of struct pt_regs */
+enum perf_event_csky_regs {
+	PERF_REG_CSKY_TLS,
+	PERF_REG_CSKY_LR,
+	PERF_REG_CSKY_PC,
+	PERF_REG_CSKY_SR,
+	PERF_REG_CSKY_SP,
+	PERF_REG_CSKY_ORIG_A0,
+	PERF_REG_CSKY_A0,
+	PERF_REG_CSKY_A1,
+	PERF_REG_CSKY_A2,
+	PERF_REG_CSKY_A3,
+	PERF_REG_CSKY_REGS0,
+	PERF_REG_CSKY_REGS1,
+	PERF_REG_CSKY_REGS2,
+	PERF_REG_CSKY_REGS3,
+	PERF_REG_CSKY_REGS4,
+	PERF_REG_CSKY_REGS5,
+	PERF_REG_CSKY_REGS6,
+	PERF_REG_CSKY_REGS7,
+	PERF_REG_CSKY_REGS8,
+	PERF_REG_CSKY_REGS9,
+#if defined(__CSKYABIV2__)
+	PERF_REG_CSKY_EXREGS0,
+	PERF_REG_CSKY_EXREGS1,
+	PERF_REG_CSKY_EXREGS2,
+	PERF_REG_CSKY_EXREGS3,
+	PERF_REG_CSKY_EXREGS4,
+	PERF_REG_CSKY_EXREGS5,
+	PERF_REG_CSKY_EXREGS6,
+	PERF_REG_CSKY_EXREGS7,
+	PERF_REG_CSKY_EXREGS8,
+	PERF_REG_CSKY_EXREGS9,
+	PERF_REG_CSKY_EXREGS10,
+	PERF_REG_CSKY_EXREGS11,
+	PERF_REG_CSKY_EXREGS12,
+	PERF_REG_CSKY_EXREGS13,
+	PERF_REG_CSKY_EXREGS14,
+	PERF_REG_CSKY_HI,
+	PERF_REG_CSKY_LO,
+	PERF_REG_CSKY_DCSR,
+#endif
+	PERF_REG_CSKY_MAX,
+};
+#endif /* _ASM_CSKY_PERF_REGS_H */
tools/perf/Makefile.config +5 -1
index fe3f97e..42985ae 100644
@@ -59,6 +59,10 @@ ifeq ($(SRCARCH),arm64)
   LIBUNWIND_LIBS = -lunwind -lunwind-aarch64
 endif
 
+ifeq ($(SRCARCH),csky)
+  NO_PERF_REGS := 0
+endif
+
 ifeq ($(ARCH),s390)
   NO_PERF_REGS := 0
   NO_SYSCALL_TABLE := 0
@@ -77,7 +81,7 @@ endif
 # Disable it on all other architectures in case libdw unwind
 # support is detected in system. Add supported architectures
 # to the check.
-ifneq ($(SRCARCH),$(filter $(SRCARCH),x86 arm arm64 powerpc s390))
+ifneq ($(SRCARCH),$(filter $(SRCARCH),x86 arm arm64 powerpc s390 csky))
   NO_LIBDW_DWARF_UNWIND := 1
 endif
 
tools/perf/arch/csky/Build new file +1 -0
new file mode 100644
index 0000000..e4e5f33
@@ -0,0 +1 @@
+perf-y += util/
tools/perf/arch/csky/Makefile new file +3 -0
new file mode 100644
index 0000000..7fbca17
@@ -0,0 +1,3 @@
+ifndef NO_DWARF
+PERF_HAVE_DWARF_REGS := 1
+endif
tools/perf/arch/csky/include/perf_regs.h new file +100 -0

100 lines not on this page · raw message

tools/perf/arch/csky/util/Build new file +2 -0
new file mode 100644
index 0000000..1160bb2
@@ -0,0 +1,2 @@
+perf-$(CONFIG_DWARF) += dwarf-regs.o
+perf-$(CONFIG_LIBDW_DWARF_UNWIND) += unwind-libdw.o
tools/perf/arch/csky/util/dwarf-regs.c new file +49 -0

49 lines not on this page · raw message

tools/perf/arch/csky/util/unwind-libdw.c new file +78 -0

78 lines not on this page · raw message

Signature · 1 line
2.7.4
author
#1 lore raw

PATCH v3 1/3 perf: use hweight64 instead of hweight_long

In reply to Mao Han

On 32-bits platform with more than 32 registers, the 64 bits mask is truncate to the lower 32 bits and the return value of hweight_long will always smaller than 32. When kernel outputs more than 32 registers, but the user perf program only counts 32, there will be a data mismatch result to overflow check fail.

CC Peter Zijlstra <[email protected]>
CC Ingo Molnar <[email protected]>
CC Arnaldo Carvalho de Melo <[email protected]>
CC Alexander Shishkin <[email protected]>
CC Jiri Olsa <[email protected]>
CC Namhyung Kim <[email protected]>
---
 tools/perf/util/evsel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/evsel.c +1 -1
index 7835e05..73c78be 100644
@@ -2322,7 +2322,7 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
 		if (data->user_regs.abi) {
 			u64 mask = evsel->attr.sample_regs_user;
 
-			sz = hweight_long(mask) * sizeof(u64);
+			sz = hweight64(mask) * sizeof(u64);
 			OVERFLOW_CHECK(array, sz, max_size);
 			data->user_regs.mask = mask;
 			data->user_regs.regs = (u64 *)array;
Signature · 1 line
2.7.4
#2 lore raw

Re: [PATCH v3 1/3] perf: use hweight64 instead of hweight_long

Em Wed, Apr 10, 2019 at 04:16:43PM +0800, Mao Han escreveu:

Quoted 23 lines

On 32-bits platform with more than 32 registers, the 64 bits mask is truncate to the lower 32 bits and the return value of hweight_long will always smaller than 32. When kernel outputs more than 32 registers, but the user perf program only counts 32, there will be a data mismatch result to overflow check fail.

CC Peter Zijlstra <[email protected]>
CC Ingo Molnar <[email protected]>
CC Arnaldo Carvalho de Melo <[email protected]>
CC Alexander Shishkin <[email protected]>
CC Jiri Olsa <[email protected]>
CC Namhyung Kim <[email protected]>
---
 tools/perf/util/evsel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/evsel.c +0 -0
index 7835e05..73c78be 100644

tools/perf/util/evsel.c int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event, · line 2328

Quoted 8 lines
tools/perf/util/evsel.c +1 -1
@@ -2322,7 +2322,7 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
 		if (data->user_regs.abi) {
 			u64 mask = evsel->attr.sample_regs_user;
 
-			sz = hweight_long(mask) * sizeof(u64);
+			sz = hweight64(mask) * sizeof(u64);
 			OVERFLOW_CHECK(array, sz, max_size);
 			data->user_regs.mask = mask;
 			data->user_regs.regs = (u64 *)array;

Later on, in the same function, perf_evsel__parse_sample() we have:

        data->intr_regs.abi = PERF_SAMPLE_REGS_ABI_NONE;
        if (type & PERF_SAMPLE_REGS_INTR) {
                OVERFLOW_CHECK_u64(array);
                data->intr_regs.abi = *array;
                array++;

                if (data->intr_regs.abi != PERF_SAMPLE_REGS_ABI_NONE) {
                        u64 mask = evsel->attr.sample_regs_intr;

                        sz = hweight_long(mask) * sizeof(u64);
                        OVERFLOW_CHECK(array, sz, max_size);
                        data->intr_regs.mask = mask;
                        data->intr_regs.regs = (u64 *)array;
                        array = (void *)array + sz;
                }
        }

You forgot to convert that one, doing it for you,

Thanks,

- Arnaldo

#3 lore raw

Re: [PATCH v3 1/3] perf: use hweight64 instead of hweight_long

Em Wed, Apr 10, 2019 at 10:08:41AM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 52 lines

Em Wed, Apr 10, 2019 at 04:16:43PM +0800, Mao Han escreveu:

Quoted 23 lines

On 32-bits platform with more than 32 registers, the 64 bits mask is truncate to the lower 32 bits and the return value of hweight_long will always smaller than 32. When kernel outputs more than 32 registers, but the user perf program only counts 32, there will be a data mismatch result to overflow check fail.

CC Peter Zijlstra <[email protected]>
CC Ingo Molnar <[email protected]>
CC Arnaldo Carvalho de Melo <[email protected]>
CC Alexander Shishkin <[email protected]>
CC Jiri Olsa <[email protected]>
CC Namhyung Kim <[email protected]>
---
 tools/perf/util/evsel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/evsel.c +0 -0
index 7835e05..73c78be 100644

tools/perf/util/evsel.c int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event, · line 2328

Quoted 8 lines
tools/perf/util/evsel.c +1 -1
@@ -2322,7 +2322,7 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
 		if (data->user_regs.abi) {
 			u64 mask = evsel->attr.sample_regs_user;
 
-			sz = hweight_long(mask) * sizeof(u64);
+			sz = hweight64(mask) * sizeof(u64);
 			OVERFLOW_CHECK(array, sz, max_size);
 			data->user_regs.mask = mask;
 			data->user_regs.regs = (u64 *)array;

Later on, in the same function, perf_evsel__parse_sample() we have:

        data->intr_regs.abi = PERF_SAMPLE_REGS_ABI_NONE;
        if (type & PERF_SAMPLE_REGS_INTR) {
                OVERFLOW_CHECK_u64(array);
                data->intr_regs.abi = *array;
                array++;

                if (data->intr_regs.abi != PERF_SAMPLE_REGS_ABI_NONE) {
                        u64 mask = evsel->attr.sample_regs_intr;

                        sz = hweight_long(mask) * sizeof(u64);
                        OVERFLOW_CHECK(array, sz, max_size);
                        data->intr_regs.mask = mask;
                        data->intr_regs.regs = (u64 *)array;
                        array = (void *)array + sz;
                }
        }
Quoted 1 line

You forgot to convert that one, doing it for you,

Also in perf_event__sample_event_size() we need to do the same thing, right?

- Arnaldo

#4 lore raw

Re: [PATCH v3 1/3] perf: use hweight64 instead of hweight_long

Em Wed, Apr 10, 2019 at 10:10:42AM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 53 lines

Em Wed, Apr 10, 2019 at 10:08:41AM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 52 lines

Em Wed, Apr 10, 2019 at 04:16:43PM +0800, Mao Han escreveu:

Quoted 23 lines

On 32-bits platform with more than 32 registers, the 64 bits mask is truncate to the lower 32 bits and the return value of hweight_long will always smaller than 32. When kernel outputs more than 32 registers, but the user perf program only counts 32, there will be a data mismatch result to overflow check fail.

CC Peter Zijlstra <[email protected]>
CC Ingo Molnar <[email protected]>
CC Arnaldo Carvalho de Melo <[email protected]>
CC Alexander Shishkin <[email protected]>
CC Jiri Olsa <[email protected]>
CC Namhyung Kim <[email protected]>
---
 tools/perf/util/evsel.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/util/evsel.c +0 -0
index 7835e05..73c78be 100644

tools/perf/util/evsel.c int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event, · line 2328

Quoted 8 lines
tools/perf/util/evsel.c +1 -1
@@ -2322,7 +2322,7 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
 		if (data->user_regs.abi) {
 			u64 mask = evsel->attr.sample_regs_user;
 
-			sz = hweight_long(mask) * sizeof(u64);
+			sz = hweight64(mask) * sizeof(u64);
 			OVERFLOW_CHECK(array, sz, max_size);
 			data->user_regs.mask = mask;
 			data->user_regs.regs = (u64 *)array;

Later on, in the same function, perf_evsel__parse_sample() we have:

        data->intr_regs.abi = PERF_SAMPLE_REGS_ABI_NONE;
        if (type & PERF_SAMPLE_REGS_INTR) {
                OVERFLOW_CHECK_u64(array);
                data->intr_regs.abi = *array;
                array++;

                if (data->intr_regs.abi != PERF_SAMPLE_REGS_ABI_NONE) {
                        u64 mask = evsel->attr.sample_regs_intr;

                        sz = hweight_long(mask) * sizeof(u64);
                        OVERFLOW_CHECK(array, sz, max_size);
                        data->intr_regs.mask = mask;
                        data->intr_regs.regs = (u64 *)array;
                        array = (void *)array + sz;
                }
        }
Quoted 4 lines
Quoted 1 line

You forgot to convert that one, doing it for you,

Also in perf_event__sample_event_size() we need to do the same thing, right?

and perf_event__synthesize_sample()

Done, resulting patch is at the end of this messages, and matches the kernel, that uses only hweight64().

I've also added Fixes tags to the patches that used hweight_long() in various places, to help with the stable trees backporting process, please consider doing it next time.

- Arnaldo

commit 21e6dfe04861c2c1b529f2759850bc62a80ca050

Author: Mao Han <[email protected]>
Date:   Wed Apr 10 16:16:43 2019 +0800

perf evsel: Use hweight64() instead of hweight_long(attr.sample_regs_user)

On 32-bits platform with more than 32 registers, the 64 bits mask is truncate to the lower 32 bits and the return value of hweight_long will always smaller than 32. When kernel outputs more than 32 registers, but the user perf program only counts 32, there will be a data mismatch result to overflow check fail.

Cc Adrian Hunter <[email protected]>
Cc Alexander Shishkin <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
Fixes 6a21c0b5c2ab ("perf tools: Add core support for sampling intr machine state regs")
Fixes d03f2170546d ("perf tools: Expand perf_event__synthesize_sample()")
Fixes 0f6a30150ca2 ("perf tools: Support user regs and stack in sample parsing")
tools/perf/util/evsel.c +6 -6
index 66d066f18b5b..966360844fff 100644
@@ -2368,7 +2368,7 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
 		if (data->user_regs.abi) {
 			u64 mask = evsel->attr.sample_regs_user;
 
-			sz = hweight_long(mask) * sizeof(u64);
+			sz = hweight64(mask) * sizeof(u64);
 			OVERFLOW_CHECK(array, sz, max_size);
 			data->user_regs.mask = mask;
 			data->user_regs.regs = (u64 *)array;
@@ -2424,7 +2424,7 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
 		if (data->intr_regs.abi != PERF_SAMPLE_REGS_ABI_NONE) {
 			u64 mask = evsel->attr.sample_regs_intr;
 
-			sz = hweight_long(mask) * sizeof(u64);
+			sz = hweight64(mask) * sizeof(u64);
 			OVERFLOW_CHECK(array, sz, max_size);
 			data->intr_regs.mask = mask;
 			data->intr_regs.regs = (u64 *)array;
@@ -2552,7 +2552,7 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
 	if (type & PERF_SAMPLE_REGS_USER) {
 		if (sample->user_regs.abi) {
 			result += sizeof(u64);
-			sz = hweight_long(sample->user_regs.mask) * sizeof(u64);
+			sz = hweight64(sample->user_regs.mask) * sizeof(u64);
 			result += sz;
 		} else {
 			result += sizeof(u64);
@@ -2580,7 +2580,7 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
 	if (type & PERF_SAMPLE_REGS_INTR) {
 		if (sample->intr_regs.abi) {
 			result += sizeof(u64);
-			sz = hweight_long(sample->intr_regs.mask) * sizeof(u64);
+			sz = hweight64(sample->intr_regs.mask) * sizeof(u64);
 			result += sz;
 		} else {
 			result += sizeof(u64);
@@ -2710,7 +2710,7 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type,
 	if (type & PERF_SAMPLE_REGS_USER) {
 		if (sample->user_regs.abi) {
 			*array++ = sample->user_regs.abi;
-			sz = hweight_long(sample->user_regs.mask) * sizeof(u64);
+			sz = hweight64(sample->user_regs.mask) * sizeof(u64);
 			memcpy(array, sample->user_regs.regs, sz);
 			array = (void *)array + sz;
 		} else {
@@ -2746,7 +2746,7 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type,
 	if (type & PERF_SAMPLE_REGS_INTR) {
 		if (sample->intr_regs.abi) {
 			*array++ = sample->intr_regs.abi;
-			sz = hweight_long(sample->intr_regs.mask) * sizeof(u64);
+			sz = hweight64(sample->intr_regs.mask) * sizeof(u64);
 			memcpy(array, sample->intr_regs.regs, sz);
 			array = (void *)array + sz;
 		} else {
author
#5 lore raw

Re: [PATCH v3 1/3] perf: use hweight64 instead of hweight_long

Quoted 7 lines · Arnaldo Carvalho de Melo
Quoted 4 lines
Quoted 1 line

You forgot to convert that one, doing it for you,

Also in perf_event__sample_event_size() we need to do the same thing, right?

and perf_event__synthesize_sample()

Quoted 9 lines

Done, resulting patch is at the end of this messages, and matches the kernel, that uses only hweight64().

I've also added Fixes tags to the patches that used hweight_long() in various places, to help with the stable trees backporting process, please consider doing it next time.

- Arnaldo

Thanks for help improving the patch and the suggestion. Tested with the new patch on C-SKY, seems work fine.

Mao Han

#6 lore raw

[tip:perf/urgent] perf evsel: Use hweight64() instead of hweight_long(attr.sample_regs_user)

In reply to Mao Han
perf evsel: Use hweight64() instead of hweight_long(attr.sample_regs_user)
Commit-ID:  21e6dfe04861c2c1b529f2759850bc62a80ca050
Gitweb:     https://git.kernel.org/tip/21e6dfe04861c2c1b529f2759850bc62a80ca050
Author:     Mao Han <[email protected]>
AuthorDate: Wed, 10 Apr 2019 16:16:43 +0800
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Wed, 10 Apr 2019 10:25:28 -0300

perf evsel: Use hweight64() instead of hweight_long(attr.sample_regs_user)

On 32-bits platform with more than 32 registers, the 64 bits mask is truncate to the lower 32 bits and the return value of hweight_long will always smaller than 32. When kernel outputs more than 32 registers, but the user perf program only counts 32, there will be a data mismatch result to overflow check fail.

Cc Adrian Hunter <[email protected]>
Cc Alexander Shishkin <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
Fixes 6a21c0b5c2ab ("perf tools: Add core support for sampling intr machine state regs")
Fixes d03f2170546d ("perf tools: Expand perf_event__synthesize_sample()")
Fixes 0f6a30150ca2 ("perf tools: Support user regs and stack in sample parsing")
---
 tools/perf/util/evsel.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
tools/perf/util/evsel.c +6 -6
index 66d066f18b5b..966360844fff 100644
@@ -2368,7 +2368,7 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
 		if (data->user_regs.abi) {
 			u64 mask = evsel->attr.sample_regs_user;
 
-			sz = hweight_long(mask) * sizeof(u64);
+			sz = hweight64(mask) * sizeof(u64);
 			OVERFLOW_CHECK(array, sz, max_size);
 			data->user_regs.mask = mask;
 			data->user_regs.regs = (u64 *)array;
@@ -2424,7 +2424,7 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
 		if (data->intr_regs.abi != PERF_SAMPLE_REGS_ABI_NONE) {
 			u64 mask = evsel->attr.sample_regs_intr;
 
-			sz = hweight_long(mask) * sizeof(u64);
+			sz = hweight64(mask) * sizeof(u64);
 			OVERFLOW_CHECK(array, sz, max_size);
 			data->intr_regs.mask = mask;
 			data->intr_regs.regs = (u64 *)array;
@@ -2552,7 +2552,7 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
 	if (type & PERF_SAMPLE_REGS_USER) {
 		if (sample->user_regs.abi) {
 			result += sizeof(u64);
-			sz = hweight_long(sample->user_regs.mask) * sizeof(u64);
+			sz = hweight64(sample->user_regs.mask) * sizeof(u64);
 			result += sz;
 		} else {
 			result += sizeof(u64);
@@ -2580,7 +2580,7 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
 	if (type & PERF_SAMPLE_REGS_INTR) {
 		if (sample->intr_regs.abi) {
 			result += sizeof(u64);
-			sz = hweight_long(sample->intr_regs.mask) * sizeof(u64);
+			sz = hweight64(sample->intr_regs.mask) * sizeof(u64);
 			result += sz;
 		} else {
 			result += sizeof(u64);
@@ -2710,7 +2710,7 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type,
 	if (type & PERF_SAMPLE_REGS_USER) {
 		if (sample->user_regs.abi) {
 			*array++ = sample->user_regs.abi;
-			sz = hweight_long(sample->user_regs.mask) * sizeof(u64);
+			sz = hweight64(sample->user_regs.mask) * sizeof(u64);
 			memcpy(array, sample->user_regs.regs, sz);
 			array = (void *)array + sz;
 		} else {
@@ -2746,7 +2746,7 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type,
 	if (type & PERF_SAMPLE_REGS_INTR) {
 		if (sample->intr_regs.abi) {
 			*array++ = sample->intr_regs.abi;
-			sz = hweight_long(sample->intr_regs.mask) * sizeof(u64);
+			sz = hweight64(sample->intr_regs.mask) * sizeof(u64);
 			memcpy(array, sample->intr_regs.regs, sz);
 			array = (void *)array + sz;
 		} else {
#7 lore raw

[tip:perf/urgent] perf evsel: Use hweight64() instead of hweight_long(attr.sample_regs_user)

In reply to Mao Han
perf evsel: Use hweight64() instead of hweight_long(attr.sample_regs_user)
Commit-ID:  3a5b64f05d7fe36dea0dde26423e3044fbacd482
Gitweb:     https://git.kernel.org/tip/3a5b64f05d7fe36dea0dde26423e3044fbacd482
Author:     Mao Han <[email protected]>
AuthorDate: Wed, 10 Apr 2019 16:16:43 +0800
Committer:  Arnaldo Carvalho de Melo <[email protected]>
CommitDate: Tue, 16 Apr 2019 11:27:53 -0300

perf evsel: Use hweight64() instead of hweight_long(attr.sample_regs_user)

On 32-bits platform with more than 32 registers, the 64 bits mask is truncate to the lower 32 bits and the return value of hweight_long will always smaller than 32. When kernel outputs more than 32 registers, but the user perf program only counts 32, there will be a data mismatch result to overflow check fail.

Cc Adrian Hunter <[email protected]>
Cc Alexander Shishkin <[email protected]>
Cc Jiri Olsa <[email protected]>
Cc Namhyung Kim <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Stephane Eranian <[email protected]>
Fixes 6a21c0b5c2ab ("perf tools: Add core support for sampling intr machine state regs")
Fixes d03f2170546d ("perf tools: Expand perf_event__synthesize_sample()")
Fixes 0f6a30150ca2 ("perf tools: Support user regs and stack in sample parsing")
---
 tools/perf/util/evsel.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
tools/perf/util/evsel.c +6 -6
index 66d066f18b5b..966360844fff 100644
@@ -2368,7 +2368,7 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
 		if (data->user_regs.abi) {
 			u64 mask = evsel->attr.sample_regs_user;
 
-			sz = hweight_long(mask) * sizeof(u64);
+			sz = hweight64(mask) * sizeof(u64);
 			OVERFLOW_CHECK(array, sz, max_size);
 			data->user_regs.mask = mask;
 			data->user_regs.regs = (u64 *)array;
@@ -2424,7 +2424,7 @@ int perf_evsel__parse_sample(struct perf_evsel *evsel, union perf_event *event,
 		if (data->intr_regs.abi != PERF_SAMPLE_REGS_ABI_NONE) {
 			u64 mask = evsel->attr.sample_regs_intr;
 
-			sz = hweight_long(mask) * sizeof(u64);
+			sz = hweight64(mask) * sizeof(u64);
 			OVERFLOW_CHECK(array, sz, max_size);
 			data->intr_regs.mask = mask;
 			data->intr_regs.regs = (u64 *)array;
@@ -2552,7 +2552,7 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
 	if (type & PERF_SAMPLE_REGS_USER) {
 		if (sample->user_regs.abi) {
 			result += sizeof(u64);
-			sz = hweight_long(sample->user_regs.mask) * sizeof(u64);
+			sz = hweight64(sample->user_regs.mask) * sizeof(u64);
 			result += sz;
 		} else {
 			result += sizeof(u64);
@@ -2580,7 +2580,7 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
 	if (type & PERF_SAMPLE_REGS_INTR) {
 		if (sample->intr_regs.abi) {
 			result += sizeof(u64);
-			sz = hweight_long(sample->intr_regs.mask) * sizeof(u64);
+			sz = hweight64(sample->intr_regs.mask) * sizeof(u64);
 			result += sz;
 		} else {
 			result += sizeof(u64);
@@ -2710,7 +2710,7 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type,
 	if (type & PERF_SAMPLE_REGS_USER) {
 		if (sample->user_regs.abi) {
 			*array++ = sample->user_regs.abi;
-			sz = hweight_long(sample->user_regs.mask) * sizeof(u64);
+			sz = hweight64(sample->user_regs.mask) * sizeof(u64);
 			memcpy(array, sample->user_regs.regs, sz);
 			array = (void *)array + sz;
 		} else {
@@ -2746,7 +2746,7 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type,
 	if (type & PERF_SAMPLE_REGS_INTR) {
 		if (sample->intr_regs.abi) {
 			*array++ = sample->intr_regs.abi;
-			sz = hweight_long(sample->intr_regs.mask) * sizeof(u64);
+			sz = hweight64(sample->intr_regs.mask) * sizeof(u64);
 			memcpy(array, sample->intr_regs.regs, sz);
 			array = (void *)array + sz;
 		} else {