On ARM the debug info is not present in the .eh_frame sections but
in .debug_frame instead.
This patch uses libunwind to load and parse the dwarf debug info from
the .debug_frame section if no .eh_frame_hdr section is found; also it
sets the hooks in the perf_regs and libunwind code for ARM.
The generated perf has been tested on ARM/OMAP4 and x86_64, using the
following commands:
perf record -g [dwarf] -- <binary>
perf report --sort symbol --call-graph --stdio
Jean Pihet (1):
perf: parse the .debug_frame section in case .eh_frame is not present
Will Deacon (2):
ARM: perf: add support for perf regs API
ARM: perf: wire up perf_regs and unwind support for ARM
arch/arm/Kconfig | 2 +
arch/arm/include/uapi/asm/Kbuild | 1 +
arch/arm/include/uapi/asm/perf_regs.h | 23 +++++++++++
arch/arm/kernel/Makefile | 1 +
arch/arm/kernel/perf_regs.c | 30 ++++++++++++++
tools/perf/arch/arm/Makefile | 3 ++
tools/perf/arch/arm/include/perf_regs.h | 54 +++++++++++++++++++++++++
tools/perf/arch/arm/util/unwind.c | 48 ++++++++++++++++++++++
tools/perf/config/Makefile | 7 +++-
tools/perf/util/unwind.c | 70 +++++++++++++++++++++++++--------
10 files changed, 221 insertions(+), 18 deletions(-)
create mode 100644 arch/arm/include/uapi/asm/perf_regs.h
create mode 100644 arch/arm/kernel/perf_regs.c
create mode 100644 tools/perf/arch/arm/include/perf_regs.h
create mode 100644 tools/perf/arch/arm/util/unwind.c
--
1.7.11.7
From: Will Deacon <[email protected]>
This patch hooks in the perf_regs and libunwind code for ARM.
Signed-off-by: Will Deacon <[email protected]>
Cc: Jean Pihet <[email protected]>
---
tools/perf/arch/arm/Makefile | 3 ++
tools/perf/arch/arm/include/perf_regs.h | 54 +++++++++++++++++++++++++++++++++
tools/perf/arch/arm/util/unwind.c | 48 +++++++++++++++++++++++++++++
tools/perf/config/Makefile | 7 +++--
4 files changed, 110 insertions(+), 2 deletions(-)
create mode 100644 tools/perf/arch/arm/include/perf_regs.h
create mode 100644 tools/perf/arch/arm/util/unwind.c
diff --git a/tools/perf/arch/arm/Makefile b/tools/perf/arch/arm/Makefile
index 15130b5..fe9b61e 100644
--- a/tools/perf/arch/arm/Makefile
+++ b/tools/perf/arch/arm/Makefile
@@ -2,3 +2,6 @@ ifndef NO_DWARF
PERF_HAVE_DWARF_REGS := 1
LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/dwarf-regs.o
endif
+ifndef NO_LIBUNWIND
+LIB_OBJS += $(OUTPUT)arch/$(ARCH)/util/unwind.o
+endif
diff --git a/tools/perf/arch/arm/include/perf_regs.h b/tools/perf/arch/arm/include/perf_regs.h
new file mode 100644
index 0000000..2a1cfde
--- /dev/null
+++ b/tools/perf/arch/arm/include/perf_regs.h
@@ -0,0 +1,54 @@
+#ifndef ARCH_PERF_REGS_H
+#define ARCH_PERF_REGS_H
+
+#include <stdlib.h>
+#include "../../util/types.h"
+#include <asm/perf_regs.h>
+
+#define PERF_REGS_MASK ((1ULL << PERF_REG_ARM_MAX) - 1)
+#define PERF_REG_IP PERF_REG_ARM_PC
+#define PERF_REG_SP PERF_REG_ARM_SP
+
+static inline const char *perf_reg_name(int id)
+{
+ switch (id) {
+ case PERF_REG_ARM_R0:
+ return "r0";
+ case PERF_REG_ARM_R1:
+ return "r1";
+ case PERF_REG_ARM_R2:
+ return "r2";
+ case PERF_REG_ARM_R3:
+ return "r3";
+ case PERF_REG_ARM_R4:
+ return "r4";
+ case PERF_REG_ARM_R5:
+ return "r5";
+ case PERF_REG_ARM_R6:
+ return "r6";
+ case PERF_REG_ARM_R7:
+ return "r7";
+ case PERF_REG_ARM_R8:
+ return "r8";
+ case PERF_REG_ARM_R9:
+ return "r9";
+ case PERF_REG_ARM_R10:
+ return "r10";
+ case PERF_REG_ARM_FP:
+ return "fp";
+ case PERF_REG_ARM_IP:
+ return "ip";
+ case PERF_REG_ARM_SP:
+ return "sp";
+ case PERF_REG_ARM_LR:
+ return "lr";
+ case PERF_REG_ARM_PC:
+ return "pc";
+ default:
+ return NULL;
+ }
+
+ return NULL;
+}
+
+#endif /* ARCH_PERF_REGS_H */
diff --git a/tools/perf/arch/arm/util/unwind.c b/tools/perf/arch/arm/util/unwind.c
new file mode 100644
index 0000000..da3dc95
--- /dev/null
+++ b/tools/perf/arch/arm/util/unwind.c
@@ -0,0 +1,48 @@
+
+#include <errno.h>
+#include <libunwind.h>
+#include "perf_regs.h"
+#include "../../util/unwind.h"
+
+int unwind__arch_reg_id(int regnum)
+{
+ switch (regnum) {
+ case UNW_ARM_R0:
+ return PERF_REG_ARM_R0;
+ case UNW_ARM_R1:
+ return PERF_REG_ARM_R1;
+ case UNW_ARM_R2:
+ return PERF_REG_ARM_R2;
+ case UNW_ARM_R3:
+ return PERF_REG_ARM_R3;
+ case UNW_ARM_R4:
+ return PERF_REG_ARM_R4;
+ case UNW_ARM_R5:
+ return PERF_REG_ARM_R5;
+ case UNW_ARM_R6:
+ return PERF_REG_ARM_R6;
+ case UNW_ARM_R7:
+ return PERF_REG_ARM_R7;
+ case UNW_ARM_R8:
+ return PERF_REG_ARM_R8;
+ case UNW_ARM_R9:
+ return PERF_REG_ARM_R9;
+ case UNW_ARM_R10:
+ return PERF_REG_ARM_R10;
+ case UNW_ARM_R11:
+ return PERF_REG_ARM_FP;
+ case UNW_ARM_R12:
+ return PERF_REG_ARM_IP;
+ case UNW_ARM_R13:
+ return PERF_REG_ARM_SP;
+ case UNW_ARM_R14:
+ return PERF_REG_ARM_LR;
+ case UNW_ARM_R15:
+ return PERF_REG_ARM_PC;
+ default:
+ pr_err("unwind: invalid reg id %d\n", regnum);
+ return -EINVAL;
+ }
+
+ return -EINVAL;
+}
diff --git a/tools/perf/config/Makefile b/tools/perf/config/Makefile
index b5d9238..7014373 100644
--- a/tools/perf/config/Makefile
+++ b/tools/perf/config/Makefile
@@ -29,6 +29,10 @@ ifeq ($(ARCH),x86_64)
NO_PERF_REGS := 0
LIBUNWIND_LIBS = -lunwind -lunwind-x86_64
endif
+ifeq ($(ARCH),arm)
+ NO_PERF_REGS := 0
+ LIBUNWIND_LIBS = -lunwind -lunwind-arm
+endif
ifeq ($(NO_PERF_REGS),0)
CFLAGS += -DHAVE_PERF_REGS
@@ -204,8 +208,7 @@ ifeq ($(call try-cc,$(SOURCE_ELF_MMAP),$(FLAGS_LIBELF),-DLIBELF_MMAP),y)
endif # try-cc
endif # NO_LIBELF
-# There's only x86 (both 32 and 64) support for CFI unwind so far
-ifneq ($(ARCH),x86)
+ifeq ($(LIBUNWIND_LIBS),)
NO_LIBUNWIND := 1
endif
--
1.7.11.7
On ARM the debug info is not present in the .eh_frame sections but
instead in .debug_frame.
Use libunwind to load and parse the debug info.
Signed-off-by: Jean Pihet <[email protected]>
---
tools/perf/util/unwind.c | 70 +++++++++++++++++++++++++++++++++++++-----------
1 file changed, 54 insertions(+), 16 deletions(-)
diff --git a/tools/perf/util/unwind.c b/tools/perf/util/unwind.c
index 958723b..5353b32 100644
--- a/tools/perf/util/unwind.c
+++ b/tools/perf/util/unwind.c
@@ -39,6 +39,14 @@ UNW_OBJ(dwarf_search_unwind_table) (unw_addr_space_t as,
#define dwarf_search_unwind_table UNW_OBJ(dwarf_search_unwind_table)
+extern int
+UNW_OBJ(dwarf_find_debug_frame) (int found, unw_dyn_info_t *di_debug,
+ unw_word_t ip, unw_word_t segbase,
+ const char *obj_name, unw_word_t start,
+ unw_word_t end);
+
+#define dwarf_find_debug_frame UNW_OBJ(dwarf_find_debug_frame)
+
#define DW_EH_PE_FORMAT_MASK 0x0f /* format of the encoded value */
#define DW_EH_PE_APPL_MASK 0x70 /* how the value is to be applied */
@@ -245,8 +253,9 @@ static int unwind_spec_ehframe(struct dso *dso, struct machine *machine,
return 0;
}
-static int read_unwind_spec(struct dso *dso, struct machine *machine,
- u64 *table_data, u64 *segbase, u64 *fde_count)
+static int read_unwind_spec_eh_frame(struct dso *dso, struct machine *machine,
+ u64 *table_data, u64 *segbase,
+ u64 *fde_count)
{
int ret = -EINVAL, fd;
u64 offset;
@@ -255,6 +264,7 @@ static int read_unwind_spec(struct dso *dso, struct machine *machine,
if (fd < 0)
return -EINVAL;
+ /* Check the .eh_frame section for unwinding info */
offset = elf_section_offset(fd, ".eh_frame_hdr");
close(fd);
@@ -263,10 +273,27 @@ static int read_unwind_spec(struct dso *dso, struct machine *machine,
table_data, segbase,
fde_count);
- /* TODO .debug_frame check if eh_frame_hdr fails */
return ret;
}
+static int read_unwind_spec_debug_frame(struct dso *dso,
+ struct machine *machine, u64 *offset)
+{
+ int fd = dso__data_fd(dso, machine);
+
+ if (fd < 0)
+ return -EINVAL;
+
+ /* Check the .debug_frame section for unwinding info */
+ *offset = elf_section_offset(fd, ".debug_frame");
+ close(fd);
+
+ if (*offset)
+ return 0;
+
+ return -EINVAL;
+}
+
static struct map *find_map(unw_word_t ip, struct unwind_info *ui)
{
struct addr_location al;
@@ -291,20 +318,31 @@ find_proc_info(unw_addr_space_t as, unw_word_t ip, unw_proc_info_t *pi,
pr_debug("unwind: find_proc_info dso %s\n", map->dso->name);
- if (read_unwind_spec(map->dso, ui->machine,
- &table_data, &segbase, &fde_count))
- return -EINVAL;
+ /* Check the .eh_frame section for unwinding info */
+ if (!read_unwind_spec_eh_frame(map->dso, ui->machine,
+ &table_data, &segbase, &fde_count)) {
+ memset(&di, 0, sizeof(di));
+ di.format = UNW_INFO_FORMAT_REMOTE_TABLE;
+ di.start_ip = map->start;
+ di.end_ip = map->end;
+ di.u.rti.segbase = map->start + segbase;
+ di.u.rti.table_data = map->start + table_data;
+ di.u.rti.table_len = fde_count * sizeof(struct table_entry)
+ / sizeof(unw_word_t);
+ return dwarf_search_unwind_table(as, ip, &di, pi,
+ need_unwind_info, arg);
+ }
+
+ /* Check the .debug_frame section for unwinding info */
+ if (!read_unwind_spec_debug_frame(map->dso, ui->machine, &segbase)) {
+ memset(&di, 0, sizeof(di));
+ dwarf_find_debug_frame(0, &di, ip, 0, map->dso->name,
+ map->start, map->end);
+ return dwarf_search_unwind_table(as, ip, &di, pi,
+ need_unwind_info, arg);
+ }
- memset(&di, 0, sizeof(di));
- di.format = UNW_INFO_FORMAT_REMOTE_TABLE;
- di.start_ip = map->start;
- di.end_ip = map->end;
- di.u.rti.segbase = map->start + segbase;
- di.u.rti.table_data = map->start + table_data;
- di.u.rti.table_len = fde_count * sizeof(struct table_entry)
- / sizeof(unw_word_t);
- return dwarf_search_unwind_table(as, ip, &di, pi,
- need_unwind_info, arg);
+ return -EINVAL;
}
static int access_fpreg(unw_addr_space_t __maybe_unused as,
--
1.7.11.7
From: Will Deacon <[email protected]>
This patch implements the functions required for the perf refs API,
allowing the perf tool to interface kernel register dumps with libunwind
in order to provide userspace backtracing.
Signed-off-by: Will Deacon <[email protected]>
---
arch/arm/Kconfig | 2 ++
arch/arm/include/uapi/asm/Kbuild | 1 +
arch/arm/include/uapi/asm/perf_regs.h | 23 +++++++++++++++++++++++
arch/arm/kernel/Makefile | 1 +
arch/arm/kernel/perf_regs.c | 30 ++++++++++++++++++++++++++++++
5 files changed, 57 insertions(+)
create mode 100644 arch/arm/include/uapi/asm/perf_regs.h
create mode 100644 arch/arm/kernel/perf_regs.c
diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 86bf7cb..642bc48 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig
@@ -49,6 +49,8 @@ config ARM
select HAVE_MEMBLOCK
select HAVE_OPROFILE if (HAVE_PERF_EVENTS)
select HAVE_PERF_EVENTS
+ select HAVE_PERF_REGS
+ select HAVE_PERF_USER_STACK_DUMP
select HAVE_REGS_AND_STACK_ACCESS_API
select HAVE_SYSCALL_TRACEPOINTS
select HAVE_UID16
diff --git a/arch/arm/include/uapi/asm/Kbuild b/arch/arm/include/uapi/asm/Kbuild
index 18d76fd..70a1c9d 100644
--- a/arch/arm/include/uapi/asm/Kbuild
+++ b/arch/arm/include/uapi/asm/Kbuild
@@ -7,6 +7,7 @@ header-y += hwcap.h
header-y += ioctls.h
header-y += kvm_para.h
header-y += mman.h
+header-y += perf_regs.h
header-y += posix_types.h
header-y += ptrace.h
header-y += setup.h
diff --git a/arch/arm/include/uapi/asm/perf_regs.h b/arch/arm/include/uapi/asm/perf_regs.h
new file mode 100644
index 0000000..ce59448
--- /dev/null
+++ b/arch/arm/include/uapi/asm/perf_regs.h
@@ -0,0 +1,23 @@
+#ifndef _ASM_ARM_PERF_REGS_H
+#define _ASM_ARM_PERF_REGS_H
+
+enum perf_event_arm_regs {
+ PERF_REG_ARM_R0,
+ PERF_REG_ARM_R1,
+ PERF_REG_ARM_R2,
+ PERF_REG_ARM_R3,
+ PERF_REG_ARM_R4,
+ PERF_REG_ARM_R5,
+ PERF_REG_ARM_R6,
+ PERF_REG_ARM_R7,
+ PERF_REG_ARM_R8,
+ PERF_REG_ARM_R9,
+ PERF_REG_ARM_R10,
+ PERF_REG_ARM_FP,
+ PERF_REG_ARM_IP,
+ PERF_REG_ARM_SP,
+ PERF_REG_ARM_LR,
+ PERF_REG_ARM_PC,
+ PERF_REG_ARM_MAX,
+};
+#endif /* _ASM_ARM_PERF_REGS_H */
diff --git a/arch/arm/kernel/Makefile b/arch/arm/kernel/Makefile
index 86d10dd..c9e05f2 100644
--- a/arch/arm/kernel/Makefile
+++ b/arch/arm/kernel/Makefile
@@ -78,6 +78,7 @@ obj-$(CONFIG_CPU_XSC3) += xscale-cp0.o
obj-$(CONFIG_CPU_MOHAWK) += xscale-cp0.o
obj-$(CONFIG_CPU_PJ4) += pj4-cp0.o
obj-$(CONFIG_IWMMXT) += iwmmxt.o
+obj-$(CONFIG_PERF_EVENTS) += perf_regs.o
obj-$(CONFIG_HW_PERF_EVENTS) += perf_event.o perf_event_cpu.o
AFLAGS_iwmmxt.o := -Wa,-mcpu=iwmmxt
obj-$(CONFIG_ARM_CPU_TOPOLOGY) += topology.o
diff --git a/arch/arm/kernel/perf_regs.c b/arch/arm/kernel/perf_regs.c
new file mode 100644
index 0000000..6e4379c
--- /dev/null
+++ b/arch/arm/kernel/perf_regs.c
@@ -0,0 +1,30 @@
+
+#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_ARM_MAX))
+ return 0;
+
+ return regs->uregs[idx];
+}
+
+#define REG_RESERVED (~((1ULL << PERF_REG_ARM_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;
+}
--
1.7.11.7
Hi Jean,
[adding Michael, since I know he was interested in this]
On Wed, Sep 04, 2013 at 07:04:14PM +0100, Jean Pihet wrote:
> On ARM the debug info is not present in the .eh_frame sections but
> instead in .debug_frame.
> Use libunwind to load and parse the debug info.
How have you tested this? Regardless of whether or not I apply this patch, I
get the same (broken/truncated) callchains for userspace in perf report.
E.g. the following stupid program (built with -O0 -g):
--->8
void bar(void)
{
int i;
for (i = 0; i < 1000000; ++i)
asm volatile("nop" ::: "memory");
}
void foo(void)
{
bar();
}
int main(void)
{
foo();
return 0;
}
8<---
Gives me an incomplete callchain:
# Overhead Command Shared Object Symbol
# ........ ........ ................. ...............................
#
0.00% unwindme unwindme [.] bar
|
--- bar
This is the same with or without your patch.
Will
Hi Will,
On Thu, Sep 5, 2013 at 2:45 PM, Will Deacon <[email protected]> wrote:
> Hi Jean,
>
> [adding Michael, since I know he was interested in this]
>
> On Wed, Sep 04, 2013 at 07:04:14PM +0100, Jean Pihet wrote:
>> On ARM the debug info is not present in the .eh_frame sections but
>> instead in .debug_frame.
>> Use libunwind to load and parse the debug info.
>
> How have you tested this? Regardless of whether or not I apply this patch, I
> get the same (broken/truncated) callchains for userspace in perf report.
Here are the commands I have been using:
perf record -g dwarf -- <binary to profile>
perf report --sort symbol --call-graph --stdio
>
> E.g. the following stupid program (built with -O0 -g):
>
> --->8
>
> void bar(void)
> {
> int i;
> for (i = 0; i < 1000000; ++i)
> asm volatile("nop" ::: "memory");
> }
>
> void foo(void)
> {
> bar();
> }
>
>
> int main(void)
> {
> foo();
> return 0;
> }
>
> 8<---
>
> Gives me an incomplete callchain:
>
> # Overhead Command Shared Object Symbol
> # ........ ........ ................. ...............................
> #
> 0.00% unwindme unwindme [.] bar
> |
> --- bar
I get the following with a simple stupid program with a long call chain:
0.57% stress_bt stress_bt [.] foo_115
|
--- foo_115
foo_114
foo_113
...
foo_92
bar
doit
main
__libc_start_main
Things to check:
- compile the binaries and libraries with -g (-dbg flavor of libs are
usually ok),
- use -g dwarf in perf record
> This is the same with or without your patch.
>
> Will
Thanks for testing!
Jean
>
> _______________________________________________
> linux-arm-kernel mailing list
> [email protected]
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
On 5 September 2013 14:45, Will Deacon <[email protected]> wrote:
> Hi Jean,
>
> [adding Michael, since I know he was interested in this]
>
> On Wed, Sep 04, 2013 at 07:04:14PM +0100, Jean Pihet wrote:
>> On ARM the debug info is not present in the .eh_frame sections but
>> instead in .debug_frame.
>> Use libunwind to load and parse the debug info.
>
> How have you tested this? Regardless of whether or not I apply this patch, I
> get the same (broken/truncated) callchains for userspace in perf report.
Same here: I have applied these patches against 3.11 running on a
Calxeda Highbank, and I never get more than a single entry for the
userland part of the call stack.
>
> E.g. the following stupid program (built with -O0 -g):
>
[...]
>
> Gives me an incomplete callchain:
>
> # Overhead Command Shared Object Symbol
> # ........ ........ ................. ...............................
> #
> 0.00% unwindme unwindme [.] bar
> |
> --- bar
>
I get exactly the same results, just bar and nothing below.
--
Ard.
On 5 September 2013 15:05, Jean Pihet <[email protected]> wrote:
[..]
> Here are the commands I have been using:
> perf record -g dwarf -- <binary to profile>
> perf report --sort symbol --call-graph --stdio
>
Ah, I failed to add the 'dwarf' after -g, however, in that case, my
perf report segfaults:
#0 locate_debug_info (as=0xb6ea9144 <local_addr_space>,
info=info@entry=0x83d6, addr=addr@entry=0, dlname=0x4a0000 <Address
0x4a0000 out of bounds>)
at dwarf/Gfind_proc_info-lsb.c:295
#1 0xb6e9a9c6 in _Uarm_dwarf_find_debug_frame (found=found@entry=0,
di_debug=di_debug@entry=0xbeff95b8, info=info@entry=0x83d6,
ip=ip@entry=0)
at dwarf/Gfind_proc_info-lsb.c:423
#2 0x0006f0e4 in find_proc_info (as=0x2c06a0, ip=33750,
pi=0xbeffa8fc, need_unwind_info=1, arg=0xbeffe530) at
util/unwind.c:339
#3 0xb6e98258 in fetch_proc_info (c=c@entry=0xbeffa4d4, ip=<optimised
out>, need_unwind_info=need_unwind_info@entry=1) at
dwarf/Gparser.c:422
#4 0xb6e99640 in uncached_dwarf_find_save_locs (c=0xbeffa4d4) at
dwarf/Gparser.c:824
#5 _Uarm_dwarf_find_save_locs (c=c@entry=0xbeffa4d4) at dwarf/Gparser.c:849
#6 0xb6e9a034 in _Uarm_dwarf_step (c=c@entry=0xbeffa4d4) at dwarf/Gstep.c:34
#7 0xb6e95182 in _Uarm_step (cursor=cursor@entry=0xbeffa4d4) at arm/Gstep.c:177
#8 0x0006ed50 in get_entries (ui=ui@entry=0xbeffe530,
cb=cb@entry=0x47599 <unwind_entry>, arg=arg@entry=0xb6b804c8) at
util/unwind.c:573
#9 0x0006f324 in unwind__get_entries (cb=cb@entry=0x47599
<unwind_entry>, arg=0xb6b804c8, machine=machine@entry=0xf0544,
thread=thread@entry=0x101860,
sample_uregs=sample_uregs@entry=65535, data=data@entry=0xbeffe740)
at util/unwind.c:608
#10 0x00049a96 in machine__resolve_callchain
(machine=machine@entry=0xf0544, evsel=evsel@entry=0xf0b98,
thread=0x101860, sample=sample@entry=0xbeffe740,
parent=parent@entry=0xbeffe644) at util/machine.c:1262
#11 0x0001a43e in perf_evsel__add_hist_entry (machine=0xf0544,
sample=0xbeffe740, al=0xbeffe640, evsel=0xf0b98) at
builtin-report.c:256
#12 process_sample_event (tool=0xbeffeb68, event=0xb6b3c600,
sample=0xbeffe740, evsel=0xf0b98, machine=0xf0544) at
builtin-report.c:335
#13 0x0004b3ca in perf_session_deliver_event
(session=session@entry=0xf0498, event=0xb6b3c600,
sample=sample@entry=0xbeffe740, tool=tool@entry=0xbeffeb68,
file_offset=file_offset@entry=161280) at util/session.c:873
#14 0x0004b8d0 in flush_sample_queue (s=s@entry=0xf0498,
tool=tool@entry=0xbeffeb68) at util/session.c:521
#15 0x0004c974 in __perf_session__process_events
(session=session@entry=0xf0498, data_offset=<optimised out>,
data_size=data_size@entry=437976, file_size=438264,
tool=tool@entry=0xbeffeb68) at util/session.c:1269
#16 0x0004cc5c in perf_session__process_events
(self=self@entry=0xf0498, tool=tool@entry=0xbeffeb68) at
util/session.c:1286
#17 0x0001b4b2 in __cmd_report (rep=0xbeffeb68) at builtin-report.c:513
#18 cmd_report (argc=0, argv=0xbefff6c8, prefix=<optimised out>) at
builtin-report.c:957
#19 0x0000d80e in run_builtin (p=p@entry=0xaa548 <commands+84>,
argc=argc@entry=2, argv=argv@entry=0xbefff6c8) at perf.c:319
#20 0x0000d28a in handle_internal_command (argv=0xbefff6c8, argc=2) at
perf.c:376
#21 run_argv (argv=0xbefff4a8, argcp=0xbefff4ac) at perf.c:420
#22 main (argc=2, argv=0xbefff6c8) at perf.c:521
--
Ard.
>>
>> E.g. the following stupid program (built with -O0 -g):
>>
>> --->8
>>
>> void bar(void)
>> {
>> int i;
>> for (i = 0; i < 1000000; ++i)
>> asm volatile("nop" ::: "memory");
>> }
>>
>> void foo(void)
>> {
>> bar();
>> }
>>
>>
>> int main(void)
>> {
>> foo();
>> return 0;
>> }
>>
>> 8<---
>>
>> Gives me an incomplete callchain:
>>
>> # Overhead Command Shared Object Symbol
>> # ........ ........ ................. ...............................
>> #
>> 0.00% unwindme unwindme [.] bar
>> |
>> --- bar
> I get the following with a simple stupid program with a long call chain:
> 0.57% stress_bt stress_bt [.] foo_115
> |
> --- foo_115
> foo_114
> foo_113
> ...
> foo_92
> bar
> doit
> main
> __libc_start_main
>
> Things to check:
> - compile the binaries and libraries with -g (-dbg flavor of libs are
> usually ok),
> - use -g dwarf in perf record
>
>
>> This is the same with or without your patch.
>>
>> Will
>
> Thanks for testing!
> Jean
>
>>
>> _______________________________________________
>> linux-arm-kernel mailing list
>> [email protected]
>> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
>
> _______________________________________________
> linux-arm-kernel mailing list
> [email protected]
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
On Thu, Sep 05, 2013 at 02:17:30PM +0100, Ard Biesheuvel wrote:
> On 5 September 2013 15:05, Jean Pihet <[email protected]> wrote:
> [..]
> > Here are the commands I have been using:
> > perf record -g dwarf -- <binary to profile>
> > perf report --sort symbol --call-graph --stdio
> >
>
> Ah, I failed to add the 'dwarf' after -g, however, in that case, my
> perf report segfaults:
Same SEGV here.
Will
On 4 September 2013 20:04, Jean Pihet <[email protected]> wrote:
> From: Will Deacon <[email protected]>
>
> This patch implements the functions required for the perf refs API,
'regs API' not 'refs API'
Regards,
Ard.
> allowing the perf tool to interface kernel register dumps with libunwind
> in order to provide userspace backtracing.
>
> Signed-off-by: Will Deacon <[email protected]>
> ---
> arch/arm/Kconfig | 2 ++
> arch/arm/include/uapi/asm/Kbuild | 1 +
> arch/arm/include/uapi/asm/perf_regs.h | 23 +++++++++++++++++++++++
> arch/arm/kernel/Makefile | 1 +
> arch/arm/kernel/perf_regs.c | 30 ++++++++++++++++++++++++++++++
> 5 files changed, 57 insertions(+)
> create mode 100644 arch/arm/include/uapi/asm/perf_regs.h
> create mode 100644 arch/arm/kernel/perf_regs.c
>
> diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
> index 86bf7cb..642bc48 100644
> --- a/arch/arm/Kconfig
> +++ b/arch/arm/Kconfig
> @@ -49,6 +49,8 @@ config ARM
> select HAVE_MEMBLOCK
> select HAVE_OPROFILE if (HAVE_PERF_EVENTS)
> select HAVE_PERF_EVENTS
> + select HAVE_PERF_REGS
> + select HAVE_PERF_USER_STACK_DUMP
> select HAVE_REGS_AND_STACK_ACCESS_API
> select HAVE_SYSCALL_TRACEPOINTS
> select HAVE_UID16
> diff --git a/arch/arm/include/uapi/asm/Kbuild b/arch/arm/include/uapi/asm/Kbuild
> index 18d76fd..70a1c9d 100644
> --- a/arch/arm/include/uapi/asm/Kbuild
> +++ b/arch/arm/include/uapi/asm/Kbuild
> @@ -7,6 +7,7 @@ header-y += hwcap.h
> header-y += ioctls.h
> header-y += kvm_para.h
> header-y += mman.h
> +header-y += perf_regs.h
> header-y += posix_types.h
> header-y += ptrace.h
> header-y += setup.h
> diff --git a/arch/arm/include/uapi/asm/perf_regs.h b/arch/arm/include/uapi/asm/perf_regs.h
> new file mode 100644
> index 0000000..ce59448
> --- /dev/null
> +++ b/arch/arm/include/uapi/asm/perf_regs.h
> @@ -0,0 +1,23 @@
> +#ifndef _ASM_ARM_PERF_REGS_H
> +#define _ASM_ARM_PERF_REGS_H
> +
> +enum perf_event_arm_regs {
> + PERF_REG_ARM_R0,
> + PERF_REG_ARM_R1,
> + PERF_REG_ARM_R2,
> + PERF_REG_ARM_R3,
> + PERF_REG_ARM_R4,
> + PERF_REG_ARM_R5,
> + PERF_REG_ARM_R6,
> + PERF_REG_ARM_R7,
> + PERF_REG_ARM_R8,
> + PERF_REG_ARM_R9,
> + PERF_REG_ARM_R10,
> + PERF_REG_ARM_FP,
> + PERF_REG_ARM_IP,
> + PERF_REG_ARM_SP,
> + PERF_REG_ARM_LR,
> + PERF_REG_ARM_PC,
> + PERF_REG_ARM_MAX,
> +};
> +#endif /* _ASM_ARM_PERF_REGS_H */
> diff --git a/arch/arm/kernel/Makefile b/arch/arm/kernel/Makefile
> index 86d10dd..c9e05f2 100644
> --- a/arch/arm/kernel/Makefile
> +++ b/arch/arm/kernel/Makefile
> @@ -78,6 +78,7 @@ obj-$(CONFIG_CPU_XSC3) += xscale-cp0.o
> obj-$(CONFIG_CPU_MOHAWK) += xscale-cp0.o
> obj-$(CONFIG_CPU_PJ4) += pj4-cp0.o
> obj-$(CONFIG_IWMMXT) += iwmmxt.o
> +obj-$(CONFIG_PERF_EVENTS) += perf_regs.o
> obj-$(CONFIG_HW_PERF_EVENTS) += perf_event.o perf_event_cpu.o
> AFLAGS_iwmmxt.o := -Wa,-mcpu=iwmmxt
> obj-$(CONFIG_ARM_CPU_TOPOLOGY) += topology.o
> diff --git a/arch/arm/kernel/perf_regs.c b/arch/arm/kernel/perf_regs.c
> new file mode 100644
> index 0000000..6e4379c
> --- /dev/null
> +++ b/arch/arm/kernel/perf_regs.c
> @@ -0,0 +1,30 @@
> +
> +#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_ARM_MAX))
> + return 0;
> +
> + return regs->uregs[idx];
> +}
> +
> +#define REG_RESERVED (~((1ULL << PERF_REG_ARM_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;
> +}
> --
> 1.7.11.7
>
>
> _______________________________________________
> linux-arm-kernel mailing list
> [email protected]
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
On 5 September 2013 15:19, Will Deacon <[email protected]> wrote:
> On Thu, Sep 05, 2013 at 02:17:30PM +0100, Ard Biesheuvel wrote:
>> On 5 September 2013 15:05, Jean Pihet <[email protected]> wrote:
>> [..]
>> > Here are the commands I have been using:
>> > perf record -g dwarf -- <binary to profile>
>> > perf report --sort symbol --call-graph --stdio
>> >
>>
>> Ah, I failed to add the 'dwarf' after -g, however, in that case, my
>> perf report segfaults:
>
> Same SEGV here.
Ouch! I never got the segfault on my side.
Here is the setup I am using:
- OMAP4 Pandaboard,
- 13.07 Ubuntu dist from http://www.linaro.org/downloads/,
- 3.11 mainline kernel + 3 patches for unwinding info,
- patched perf tool,
- dbg flavor of the libs installed. Without them installed I do not
get the address resolved but no segfaults
Can you check if the linked libraries contain the .debug_frame
sections? (ldd; readelf -S /lib/arm-linux-gnueabihf/libc-2.17.so|grep
debug_frame; readelf -S
/usr/lib/debug/lib/arm-linux-gnueabihf/libc-2.17.so|grep debug_frame)?
>
> Will
Jean
Hi,
On 5 September 2013 15:33, Jean Pihet <[email protected]> wrote:
> On 5 September 2013 15:19, Will Deacon <[email protected]> wrote:
>> On Thu, Sep 05, 2013 at 02:17:30PM +0100, Ard Biesheuvel wrote:
>>> On 5 September 2013 15:05, Jean Pihet <[email protected]> wrote:
>>> [..]
>>> > Here are the commands I have been using:
>>> > perf record -g dwarf -- <binary to profile>
>>> > perf report --sort symbol --call-graph --stdio
>>> >
>>>
>>> Ah, I failed to add the 'dwarf' after -g, however, in that case, my
>>> perf report segfaults:
>>
>> Same SEGV here.
> Ouch! I never got the segfault on my side.
The segfault is caused by libunwind. With the latest version [1],
dwarf unwinding works fine.
Investigation on-going, more to come!
[1] git://git.sv.gnu.org/libunwind.git
Regards,
Jean
>
> Here is the setup I am using:
> - OMAP4 Pandaboard,
> - 13.07 Ubuntu dist from http://www.linaro.org/downloads/,
> - 3.11 mainline kernel + 3 patches for unwinding info,
> - patched perf tool,
> - dbg flavor of the libs installed. Without them installed I do not
> get the address resolved but no segfaults
>
> Can you check if the linked libraries contain the .debug_frame
> sections? (ldd; readelf -S /lib/arm-linux-gnueabihf/libc-2.17.so|grep
> debug_frame; readelf -S
> /usr/lib/debug/lib/arm-linux-gnueabihf/libc-2.17.so|grep debug_frame)?
>
>>
>> Will
>
> Jean
On 5 September 2013 17:29, Jean Pihet <[email protected]> wrote:
[...]
> The segfault is caused by libunwind. With the latest version [1],
> dwarf unwinding works fine.
> Investigation on-going, more to come!
>
> [1] git://git.sv.gnu.org/libunwind.git
>
I managed to get it to work by upgrading to Ubuntu Saucy's libunwind8
on a Ubuntu Raring system (Calxeda Highbank)
FWIW, for the series
Tested-by: Ard Biesheuvel <[email protected]> # on
ARM/Highbank/UbuntuSaucy
--
Ard.
On Wed, Sep 04, 2013 at 08:04:14PM +0200, Jean Pihet wrote:
> On ARM the debug info is not present in the .eh_frame sections but
> instead in .debug_frame.
> Use libunwind to load and parse the debug info.
hum, cannot make final link:
$ make LIBUNWIND_DIR=/opt/libunwind/
CHK -fstack-protector-all
CHK -Wstack-protector
CHK -Wvolatile-register-var
CHK -D_FORTIFY_SOURCE=2
CHK bionic
CHK libelf
CHK libdw
CHK -DLIBELF_MMAP
CHK -DLIBELF_MMAP
CHK libunwind
CHK libaudit
...
make[1]: `liblk.a' is up to date.
SUBDIR /home/jolsa/linux-perf/tools/lib/traceevent/
LINK perf
libperf.a(unwind.o): In function `find_proc_info':
/home/jolsa/linux-perf/tools/perf/util/unwind.c:339: undefined reference to `_Ux86_64_dwarf_find_debug_frame'
collect2: ld returned 1 exit status
make: *** [perf] Error 1
I'm using the latest code from git://git.sv.gnu.org/libunwind.git
Looks like dwarf_find_debug_frame is not exported, although
it looks like it is based on what I see in libunwind sources ;-)
What did I miss?
Also few typo comments below..
thanks,
jirka
>
> Signed-off-by: Jean Pihet <[email protected]>
> ---
> tools/perf/util/unwind.c | 70 +++++++++++++++++++++++++++++++++++++-----------
> 1 file changed, 54 insertions(+), 16 deletions(-)
>
> diff --git a/tools/perf/util/unwind.c b/tools/perf/util/unwind.c
> index 958723b..5353b32 100644
> --- a/tools/perf/util/unwind.c
> +++ b/tools/perf/util/unwind.c
> @@ -39,6 +39,14 @@ UNW_OBJ(dwarf_search_unwind_table) (unw_addr_space_t as,
>
> #define dwarf_search_unwind_table UNW_OBJ(dwarf_search_unwind_table)
>
> +extern int
> +UNW_OBJ(dwarf_find_debug_frame) (int found, unw_dyn_info_t *di_debug,
> + unw_word_t ip, unw_word_t segbase,
> + const char *obj_name, unw_word_t start,
> + unw_word_t end);
> +
> +#define dwarf_find_debug_frame UNW_OBJ(dwarf_find_debug_frame)
> +
> #define DW_EH_PE_FORMAT_MASK 0x0f /* format of the encoded value */
> #define DW_EH_PE_APPL_MASK 0x70 /* how the value is to be applied */
>
> @@ -245,8 +253,9 @@ static int unwind_spec_ehframe(struct dso *dso, struct machine *machine,
> return 0;
> }
>
> -static int read_unwind_spec(struct dso *dso, struct machine *machine,
> - u64 *table_data, u64 *segbase, u64 *fde_count)
> +static int read_unwind_spec_eh_frame(struct dso *dso, struct machine *machine,
> + u64 *table_data, u64 *segbase,
> + u64 *fde_count)
> {
> int ret = -EINVAL, fd;
> u64 offset;
> @@ -255,6 +264,7 @@ static int read_unwind_spec(struct dso *dso, struct machine *machine,
> if (fd < 0)
> return -EINVAL;
>
> + /* Check the .eh_frame section for unwinding info */
> offset = elf_section_offset(fd, ".eh_frame_hdr");
> close(fd);
>
> @@ -263,10 +273,27 @@ static int read_unwind_spec(struct dso *dso, struct machine *machine,
> table_data, segbase,
> fde_count);
>
> - /* TODO .debug_frame check if eh_frame_hdr fails */
> return ret;
> }
>
> +static int read_unwind_spec_debug_frame(struct dso *dso,
> + struct machine *machine, u64 *offset)
> +{
some strange formatting issue ^^^ ;-)
> + int fd = dso__data_fd(dso, machine);
> +
> + if (fd < 0)
> + return -EINVAL;
> +
> + /* Check the .debug_frame section for unwinding info */
> + *offset = elf_section_offset(fd, ".debug_frame");
> + close(fd);
> +
> + if (*offset)
> + return 0;
> +
> + return -EINVAL;
> +}
> +
> static struct map *find_map(unw_word_t ip, struct unwind_info *ui)
> {
> struct addr_location al;
> @@ -291,20 +318,31 @@ find_proc_info(unw_addr_space_t as, unw_word_t ip, unw_proc_info_t *pi,
>
> pr_debug("unwind: find_proc_info dso %s\n", map->dso->name);
>
> - if (read_unwind_spec(map->dso, ui->machine,
> - &table_data, &segbase, &fde_count))
> - return -EINVAL;
> + /* Check the .eh_frame section for unwinding info */
> + if (!read_unwind_spec_eh_frame(map->dso, ui->machine,
> + &table_data, &segbase, &fde_count)) {
ditto ^^^
> + memset(&di, 0, sizeof(di));
> + di.format = UNW_INFO_FORMAT_REMOTE_TABLE;
> + di.start_ip = map->start;
> + di.end_ip = map->end;
> + di.u.rti.segbase = map->start + segbase;
> + di.u.rti.table_data = map->start + table_data;
> + di.u.rti.table_len = fde_count * sizeof(struct table_entry)
> + / sizeof(unw_word_t);
> + return dwarf_search_unwind_table(as, ip, &di, pi,
> + need_unwind_info, arg);
ditto ^^^
> + }
> +
> + /* Check the .debug_frame section for unwinding info */
> + if (!read_unwind_spec_debug_frame(map->dso, ui->machine, &segbase)) {
> + memset(&di, 0, sizeof(di));
> + dwarf_find_debug_frame(0, &di, ip, 0, map->dso->name,
> + map->start, map->end);
> + return dwarf_search_unwind_table(as, ip, &di, pi,
> + need_unwind_info, arg);
ditto ^^^
> + }
>
> - memset(&di, 0, sizeof(di));
> - di.format = UNW_INFO_FORMAT_REMOTE_TABLE;
> - di.start_ip = map->start;
> - di.end_ip = map->end;
> - di.u.rti.segbase = map->start + segbase;
> - di.u.rti.table_data = map->start + table_data;
> - di.u.rti.table_len = fde_count * sizeof(struct table_entry)
> - / sizeof(unw_word_t);
> - return dwarf_search_unwind_table(as, ip, &di, pi,
> - need_unwind_info, arg);
> + return -EINVAL;
> }
>
> static int access_fpreg(unw_addr_space_t __maybe_unused as,
> --
> 1.7.11.7
>
Hi Jiri,
On 5 September 2013 18:30, Jiri Olsa <[email protected]> wrote:
> On Wed, Sep 04, 2013 at 08:04:14PM +0200, Jean Pihet wrote:
>> On ARM the debug info is not present in the .eh_frame sections but
>> instead in .debug_frame.
>> Use libunwind to load and parse the debug info.
>
> hum, cannot make final link:
>
> $ make LIBUNWIND_DIR=/opt/libunwind/
> CHK -fstack-protector-all
> CHK -Wstack-protector
> CHK -Wvolatile-register-var
> CHK -D_FORTIFY_SOURCE=2
> CHK bionic
> CHK libelf
> CHK libdw
> CHK -DLIBELF_MMAP
> CHK -DLIBELF_MMAP
> CHK libunwind
> CHK libaudit
>
> ...
>
> make[1]: `liblk.a' is up to date.
> SUBDIR /home/jolsa/linux-perf/tools/lib/traceevent/
> LINK perf
> libperf.a(unwind.o): In function `find_proc_info':
> /home/jolsa/linux-perf/tools/perf/util/unwind.c:339: undefined reference to `_Ux86_64_dwarf_find_debug_frame'
> collect2: ld returned 1 exit status
> make: *** [perf] Error 1
>
>
> I'm using the latest code from git://git.sv.gnu.org/libunwind.git
>
> Looks like dwarf_find_debug_frame is not exported, although
> it looks like it is based on what I see in libunwind sources ;-)
>
> What did I miss?
Weird, I do not have the error on x86_64. I am investigating this and
will come back to you asap.
>
> Also few typo comments below..
>
> thanks,
> jirka
>
>>
>> Signed-off-by: Jean Pihet <[email protected]>
>> ---
>> tools/perf/util/unwind.c | 70 +++++++++++++++++++++++++++++++++++++-----------
>> 1 file changed, 54 insertions(+), 16 deletions(-)
>>
>> diff --git a/tools/perf/util/unwind.c b/tools/perf/util/unwind.c
>> index 958723b..5353b32 100644
>> --- a/tools/perf/util/unwind.c
>> +++ b/tools/perf/util/unwind.c
>> @@ -39,6 +39,14 @@ UNW_OBJ(dwarf_search_unwind_table) (unw_addr_space_t as,
>>
>> #define dwarf_search_unwind_table UNW_OBJ(dwarf_search_unwind_table)
>>
>> +extern int
>> +UNW_OBJ(dwarf_find_debug_frame) (int found, unw_dyn_info_t *di_debug,
>> + unw_word_t ip, unw_word_t segbase,
>> + const char *obj_name, unw_word_t start,
>> + unw_word_t end);
>> +
>> +#define dwarf_find_debug_frame UNW_OBJ(dwarf_find_debug_frame)
>> +
>> #define DW_EH_PE_FORMAT_MASK 0x0f /* format of the encoded value */
>> #define DW_EH_PE_APPL_MASK 0x70 /* how the value is to be applied */
>>
>> @@ -245,8 +253,9 @@ static int unwind_spec_ehframe(struct dso *dso, struct machine *machine,
>> return 0;
>> }
>>
>> -static int read_unwind_spec(struct dso *dso, struct machine *machine,
>> - u64 *table_data, u64 *segbase, u64 *fde_count)
>> +static int read_unwind_spec_eh_frame(struct dso *dso, struct machine *machine,
>> + u64 *table_data, u64 *segbase,
>> + u64 *fde_count)
>> {
>> int ret = -EINVAL, fd;
>> u64 offset;
>> @@ -255,6 +264,7 @@ static int read_unwind_spec(struct dso *dso, struct machine *machine,
>> if (fd < 0)
>> return -EINVAL;
>>
>> + /* Check the .eh_frame section for unwinding info */
>> offset = elf_section_offset(fd, ".eh_frame_hdr");
>> close(fd);
>>
>> @@ -263,10 +273,27 @@ static int read_unwind_spec(struct dso *dso, struct machine *machine,
>> table_data, segbase,
>> fde_count);
>>
>> - /* TODO .debug_frame check if eh_frame_hdr fails */
>> return ret;
>> }
>>
>> +static int read_unwind_spec_debug_frame(struct dso *dso,
>> + struct machine *machine, u64 *offset)
>> +{
>
> some strange formatting issue ^^^ ;-)
I am using 4-spaces tabs. scripts/checkpatch.pl reported it to me in
the form of too long lines.
Should I change the code to 8-spaces tabs?
Thx!
Jean
>
>
>> + int fd = dso__data_fd(dso, machine);
>> +
>> + if (fd < 0)
>> + return -EINVAL;
>> +
>> + /* Check the .debug_frame section for unwinding info */
>> + *offset = elf_section_offset(fd, ".debug_frame");
>> + close(fd);
>> +
>> + if (*offset)
>> + return 0;
>> +
>> + return -EINVAL;
>> +}
>> +
>> static struct map *find_map(unw_word_t ip, struct unwind_info *ui)
>> {
>> struct addr_location al;
>> @@ -291,20 +318,31 @@ find_proc_info(unw_addr_space_t as, unw_word_t ip, unw_proc_info_t *pi,
>>
>> pr_debug("unwind: find_proc_info dso %s\n", map->dso->name);
>>
>> - if (read_unwind_spec(map->dso, ui->machine,
>> - &table_data, &segbase, &fde_count))
>> - return -EINVAL;
>> + /* Check the .eh_frame section for unwinding info */
>> + if (!read_unwind_spec_eh_frame(map->dso, ui->machine,
>> + &table_data, &segbase, &fde_count)) {
>
> ditto ^^^
>
>> + memset(&di, 0, sizeof(di));
>> + di.format = UNW_INFO_FORMAT_REMOTE_TABLE;
>> + di.start_ip = map->start;
>> + di.end_ip = map->end;
>> + di.u.rti.segbase = map->start + segbase;
>> + di.u.rti.table_data = map->start + table_data;
>> + di.u.rti.table_len = fde_count * sizeof(struct table_entry)
>> + / sizeof(unw_word_t);
>> + return dwarf_search_unwind_table(as, ip, &di, pi,
>> + need_unwind_info, arg);
>
> ditto ^^^
>
>> + }
>> +
>> + /* Check the .debug_frame section for unwinding info */
>> + if (!read_unwind_spec_debug_frame(map->dso, ui->machine, &segbase)) {
>> + memset(&di, 0, sizeof(di));
>> + dwarf_find_debug_frame(0, &di, ip, 0, map->dso->name,
>> + map->start, map->end);
>> + return dwarf_search_unwind_table(as, ip, &di, pi,
>> + need_unwind_info, arg);
>
> ditto ^^^
>
>> + }
>>
>> - memset(&di, 0, sizeof(di));
>> - di.format = UNW_INFO_FORMAT_REMOTE_TABLE;
>> - di.start_ip = map->start;
>> - di.end_ip = map->end;
>> - di.u.rti.segbase = map->start + segbase;
>> - di.u.rti.table_data = map->start + table_data;
>> - di.u.rti.table_len = fde_count * sizeof(struct table_entry)
>> - / sizeof(unw_word_t);
>> - return dwarf_search_unwind_table(as, ip, &di, pi,
>> - need_unwind_info, arg);
>> + return -EINVAL;
>> }
>>
>> static int access_fpreg(unw_addr_space_t __maybe_unused as,
>> --
>> 1.7.11.7
>>
Hi Jiri,
On 5 September 2013 18:30, Jiri Olsa <[email protected]> wrote:
> On Wed, Sep 04, 2013 at 08:04:14PM +0200, Jean Pihet wrote:
>> On ARM the debug info is not present in the .eh_frame sections but
>> instead in .debug_frame.
>> Use libunwind to load and parse the debug info.
>
> hum, cannot make final link:
>
> $ make LIBUNWIND_DIR=/opt/libunwind/
> CHK -fstack-protector-all
> CHK -Wstack-protector
> CHK -Wvolatile-register-var
> CHK -D_FORTIFY_SOURCE=2
> CHK bionic
> CHK libelf
> CHK libdw
> CHK -DLIBELF_MMAP
> CHK -DLIBELF_MMAP
> CHK libunwind
> CHK libaudit
>
> ...
>
> make[1]: `liblk.a' is up to date.
> SUBDIR /home/jolsa/linux-perf/tools/lib/traceevent/
> LINK perf
> libperf.a(unwind.o): In function `find_proc_info':
> /home/jolsa/linux-perf/tools/perf/util/unwind.c:339: undefined reference to `_Ux86_64_dwarf_find_debug_frame'
> collect2: ld returned 1 exit status
> make: *** [perf] Error 1
>
>
> I'm using the latest code from git://git.sv.gnu.org/libunwind.git
>
> Looks like dwarf_find_debug_frame is not exported, although
> it looks like it is based on what I see in libunwind sources ;-)
>
> What did I miss?
libunwind needs to be configured with --enable-debug-frame for the
debug_frame code to be included in the lib.
On ARM the flag is always set while it isn't on x86. Here is the
culprit below (lines from libunwind configure).
Should that be changed in configure along with the changes in tools/perf?
{ $as_echo "$as_me:${as_lineno-$LINENO}: checking whether to load
.debug_frame sections" >&5
$as_echo_n "checking whether to load .debug_frame sections... " >&6; }
# Check whether --enable-debug_frame was given.
if test "${enable_debug_frame+set}" = set; then :
enableval=$enable_debug_frame;
else
case "${target_arch}" in
(arm) enable_debug_frame=yes;;
(*) enable_debug_frame=no;;
esac
fi
if test x$enable_debug_frame = xyes; then
$as_echo "#define CONFIG_DEBUG_FRAME /**/" >>confdefs.h
fi
Regards,
Jean
>
> Also few typo comments below..
>
> thanks,
> jirka
>
>>
>> Signed-off-by: Jean Pihet <[email protected]>
>> ---
>> tools/perf/util/unwind.c | 70 +++++++++++++++++++++++++++++++++++++-----------
>> 1 file changed, 54 insertions(+), 16 deletions(-)
>>
>> diff --git a/tools/perf/util/unwind.c b/tools/perf/util/unwind.c
>> index 958723b..5353b32 100644
>> --- a/tools/perf/util/unwind.c
>> +++ b/tools/perf/util/unwind.c
>> @@ -39,6 +39,14 @@ UNW_OBJ(dwarf_search_unwind_table) (unw_addr_space_t as,
>>
>> #define dwarf_search_unwind_table UNW_OBJ(dwarf_search_unwind_table)
>>
>> +extern int
>> +UNW_OBJ(dwarf_find_debug_frame) (int found, unw_dyn_info_t *di_debug,
>> + unw_word_t ip, unw_word_t segbase,
>> + const char *obj_name, unw_word_t start,
>> + unw_word_t end);
>> +
>> +#define dwarf_find_debug_frame UNW_OBJ(dwarf_find_debug_frame)
>> +
>> #define DW_EH_PE_FORMAT_MASK 0x0f /* format of the encoded value */
>> #define DW_EH_PE_APPL_MASK 0x70 /* how the value is to be applied */
>>
>> @@ -245,8 +253,9 @@ static int unwind_spec_ehframe(struct dso *dso, struct machine *machine,
>> return 0;
>> }
>>
>> -static int read_unwind_spec(struct dso *dso, struct machine *machine,
>> - u64 *table_data, u64 *segbase, u64 *fde_count)
>> +static int read_unwind_spec_eh_frame(struct dso *dso, struct machine *machine,
>> + u64 *table_data, u64 *segbase,
>> + u64 *fde_count)
>> {
>> int ret = -EINVAL, fd;
>> u64 offset;
>> @@ -255,6 +264,7 @@ static int read_unwind_spec(struct dso *dso, struct machine *machine,
>> if (fd < 0)
>> return -EINVAL;
>>
>> + /* Check the .eh_frame section for unwinding info */
>> offset = elf_section_offset(fd, ".eh_frame_hdr");
>> close(fd);
>>
>> @@ -263,10 +273,27 @@ static int read_unwind_spec(struct dso *dso, struct machine *machine,
>> table_data, segbase,
>> fde_count);
>>
>> - /* TODO .debug_frame check if eh_frame_hdr fails */
>> return ret;
>> }
>>
>> +static int read_unwind_spec_debug_frame(struct dso *dso,
>> + struct machine *machine, u64 *offset)
>> +{
>
> some strange formatting issue ^^^ ;-)
>
>
>> + int fd = dso__data_fd(dso, machine);
>> +
>> + if (fd < 0)
>> + return -EINVAL;
>> +
>> + /* Check the .debug_frame section for unwinding info */
>> + *offset = elf_section_offset(fd, ".debug_frame");
>> + close(fd);
>> +
>> + if (*offset)
>> + return 0;
>> +
>> + return -EINVAL;
>> +}
>> +
>> static struct map *find_map(unw_word_t ip, struct unwind_info *ui)
>> {
>> struct addr_location al;
>> @@ -291,20 +318,31 @@ find_proc_info(unw_addr_space_t as, unw_word_t ip, unw_proc_info_t *pi,
>>
>> pr_debug("unwind: find_proc_info dso %s\n", map->dso->name);
>>
>> - if (read_unwind_spec(map->dso, ui->machine,
>> - &table_data, &segbase, &fde_count))
>> - return -EINVAL;
>> + /* Check the .eh_frame section for unwinding info */
>> + if (!read_unwind_spec_eh_frame(map->dso, ui->machine,
>> + &table_data, &segbase, &fde_count)) {
>
> ditto ^^^
>
>> + memset(&di, 0, sizeof(di));
>> + di.format = UNW_INFO_FORMAT_REMOTE_TABLE;
>> + di.start_ip = map->start;
>> + di.end_ip = map->end;
>> + di.u.rti.segbase = map->start + segbase;
>> + di.u.rti.table_data = map->start + table_data;
>> + di.u.rti.table_len = fde_count * sizeof(struct table_entry)
>> + / sizeof(unw_word_t);
>> + return dwarf_search_unwind_table(as, ip, &di, pi,
>> + need_unwind_info, arg);
>
> ditto ^^^
>
>> + }
>> +
>> + /* Check the .debug_frame section for unwinding info */
>> + if (!read_unwind_spec_debug_frame(map->dso, ui->machine, &segbase)) {
>> + memset(&di, 0, sizeof(di));
>> + dwarf_find_debug_frame(0, &di, ip, 0, map->dso->name,
>> + map->start, map->end);
>> + return dwarf_search_unwind_table(as, ip, &di, pi,
>> + need_unwind_info, arg);
>
> ditto ^^^
>
>> + }
>>
>> - memset(&di, 0, sizeof(di));
>> - di.format = UNW_INFO_FORMAT_REMOTE_TABLE;
>> - di.start_ip = map->start;
>> - di.end_ip = map->end;
>> - di.u.rti.segbase = map->start + segbase;
>> - di.u.rti.table_data = map->start + table_data;
>> - di.u.rti.table_len = fde_count * sizeof(struct table_entry)
>> - / sizeof(unw_word_t);
>> - return dwarf_search_unwind_table(as, ip, &di, pi,
>> - need_unwind_info, arg);
>> + return -EINVAL;
>> }
>>
>> static int access_fpreg(unw_addr_space_t __maybe_unused as,
>> --
>> 1.7.11.7
>>
On Thu, Sep 05, 2013 at 06:49:13PM +0200, Jean Pihet wrote:
> Hi Jiri,
>
> On 5 September 2013 18:30, Jiri Olsa <[email protected]> wrote:
> > On Wed, Sep 04, 2013 at 08:04:14PM +0200, Jean Pihet wrote:
> >> On ARM the debug info is not present in the .eh_frame sections but
> >> instead in .debug_frame.
> >> Use libunwind to load and parse the debug info.
> >
> > hum, cannot make final link:
> >
> > $ make LIBUNWIND_DIR=/opt/libunwind/
> > CHK -fstack-protector-all
> > CHK -Wstack-protector
> > CHK -Wvolatile-register-var
> > CHK -D_FORTIFY_SOURCE=2
> > CHK bionic
> > CHK libelf
> > CHK libdw
> > CHK -DLIBELF_MMAP
> > CHK -DLIBELF_MMAP
> > CHK libunwind
> > CHK libaudit
> >
> > ...
> >
> > make[1]: `liblk.a' is up to date.
> > SUBDIR /home/jolsa/linux-perf/tools/lib/traceevent/
> > LINK perf
> > libperf.a(unwind.o): In function `find_proc_info':
> > /home/jolsa/linux-perf/tools/perf/util/unwind.c:339: undefined reference to `_Ux86_64_dwarf_find_debug_frame'
> > collect2: ld returned 1 exit status
> > make: *** [perf] Error 1
> >
> >
> > I'm using the latest code from git://git.sv.gnu.org/libunwind.git
> >
> > Looks like dwarf_find_debug_frame is not exported, although
> > it looks like it is based on what I see in libunwind sources ;-)
> >
> > What did I miss?
> Weird, I do not have the error on x86_64. I am investigating this and
> will come back to you asap.
>
> >
> > Also few typo comments below..
> >
> > thanks,
> > jirka
> >
> >>
> >> Signed-off-by: Jean Pihet <[email protected]>
> >> ---
> >> tools/perf/util/unwind.c | 70 +++++++++++++++++++++++++++++++++++++-----------
> >> 1 file changed, 54 insertions(+), 16 deletions(-)
> >>
> >> diff --git a/tools/perf/util/unwind.c b/tools/perf/util/unwind.c
> >> index 958723b..5353b32 100644
> >> --- a/tools/perf/util/unwind.c
> >> +++ b/tools/perf/util/unwind.c
> >> @@ -39,6 +39,14 @@ UNW_OBJ(dwarf_search_unwind_table) (unw_addr_space_t as,
> >>
> >> #define dwarf_search_unwind_table UNW_OBJ(dwarf_search_unwind_table)
> >>
> >> +extern int
> >> +UNW_OBJ(dwarf_find_debug_frame) (int found, unw_dyn_info_t *di_debug,
> >> + unw_word_t ip, unw_word_t segbase,
> >> + const char *obj_name, unw_word_t start,
> >> + unw_word_t end);
> >> +
> >> +#define dwarf_find_debug_frame UNW_OBJ(dwarf_find_debug_frame)
> >> +
> >> #define DW_EH_PE_FORMAT_MASK 0x0f /* format of the encoded value */
> >> #define DW_EH_PE_APPL_MASK 0x70 /* how the value is to be applied */
> >>
> >> @@ -245,8 +253,9 @@ static int unwind_spec_ehframe(struct dso *dso, struct machine *machine,
> >> return 0;
> >> }
> >>
> >> -static int read_unwind_spec(struct dso *dso, struct machine *machine,
> >> - u64 *table_data, u64 *segbase, u64 *fde_count)
> >> +static int read_unwind_spec_eh_frame(struct dso *dso, struct machine *machine,
> >> + u64 *table_data, u64 *segbase,
> >> + u64 *fde_count)
> >> {
> >> int ret = -EINVAL, fd;
> >> u64 offset;
> >> @@ -255,6 +264,7 @@ static int read_unwind_spec(struct dso *dso, struct machine *machine,
> >> if (fd < 0)
> >> return -EINVAL;
> >>
> >> + /* Check the .eh_frame section for unwinding info */
> >> offset = elf_section_offset(fd, ".eh_frame_hdr");
> >> close(fd);
> >>
> >> @@ -263,10 +273,27 @@ static int read_unwind_spec(struct dso *dso, struct machine *machine,
> >> table_data, segbase,
> >> fde_count);
> >>
> >> - /* TODO .debug_frame check if eh_frame_hdr fails */
> >> return ret;
> >> }
> >>
> >> +static int read_unwind_spec_debug_frame(struct dso *dso,
> >> + struct machine *machine, u64 *offset)
> >> +{
> >
> > some strange formatting issue ^^^ ;-)
> I am using 4-spaces tabs. scripts/checkpatch.pl reported it to me in
> the form of too long lines.
> Should I change the code to 8-spaces tabs?
>
right, if I set tabs to 4 space then it's ok
so I guess 8 is globally agreed ;-)
jirka
On Fri, Sep 06, 2013 at 11:31:17AM +0200, Jean Pihet wrote:
> Hi Jiri,
>
> On 5 September 2013 18:30, Jiri Olsa <[email protected]> wrote:
> > On Wed, Sep 04, 2013 at 08:04:14PM +0200, Jean Pihet wrote:
> >> On ARM the debug info is not present in the .eh_frame sections but
> >> instead in .debug_frame.
> >> Use libunwind to load and parse the debug info.
> >
> > hum, cannot make final link:
> >
> > $ make LIBUNWIND_DIR=/opt/libunwind/
> > CHK -fstack-protector-all
> > CHK -Wstack-protector
> > CHK -Wvolatile-register-var
> > CHK -D_FORTIFY_SOURCE=2
> > CHK bionic
> > CHK libelf
> > CHK libdw
> > CHK -DLIBELF_MMAP
> > CHK -DLIBELF_MMAP
> > CHK libunwind
> > CHK libaudit
> >
> > ...
> >
> > make[1]: `liblk.a' is up to date.
> > SUBDIR /home/jolsa/linux-perf/tools/lib/traceevent/
> > LINK perf
> > libperf.a(unwind.o): In function `find_proc_info':
> > /home/jolsa/linux-perf/tools/perf/util/unwind.c:339: undefined reference to `_Ux86_64_dwarf_find_debug_frame'
> > collect2: ld returned 1 exit status
> > make: *** [perf] Error 1
> >
> >
> > I'm using the latest code from git://git.sv.gnu.org/libunwind.git
> >
> > Looks like dwarf_find_debug_frame is not exported, although
> > it looks like it is based on what I see in libunwind sources ;-)
> >
> > What did I miss?
> libunwind needs to be configured with --enable-debug-frame for the
> debug_frame code to be included in the lib.
> On ARM the flag is always set while it isn't on x86. Here is the
> culprit below (lines from libunwind configure).
yay, thats it!
>
> Should that be changed in configure along with the changes in tools/perf?
I guess it's ok
Other than that tabs misformating the 'perf tool' change looks ok.
I tested the '.eh_frame' code and it's still working. Once
I figure out how to create a x86 binary with .debug_frame
data I'll test the code itself ;-)
If you could think of any automated testcase for this that could
be added under 'tests' that'd be nice (not necessarily)
thanks,
jirka
On 6 September 2013 12:17, Jiri Olsa <[email protected]> wrote:
> On Fri, Sep 06, 2013 at 11:31:17AM +0200, Jean Pihet wrote:
>> Hi Jiri,
>>
>> On 5 September 2013 18:30, Jiri Olsa <[email protected]> wrote:
>> > On Wed, Sep 04, 2013 at 08:04:14PM +0200, Jean Pihet wrote:
>> >> On ARM the debug info is not present in the .eh_frame sections but
>> >> instead in .debug_frame.
>> >> Use libunwind to load and parse the debug info.
>> >
>> > hum, cannot make final link:
>> >
>> > $ make LIBUNWIND_DIR=/opt/libunwind/
>> > CHK -fstack-protector-all
>> > CHK -Wstack-protector
>> > CHK -Wvolatile-register-var
>> > CHK -D_FORTIFY_SOURCE=2
>> > CHK bionic
>> > CHK libelf
>> > CHK libdw
>> > CHK -DLIBELF_MMAP
>> > CHK -DLIBELF_MMAP
>> > CHK libunwind
>> > CHK libaudit
>> >
>> > ...
>> >
>> > make[1]: `liblk.a' is up to date.
>> > SUBDIR /home/jolsa/linux-perf/tools/lib/traceevent/
>> > LINK perf
>> > libperf.a(unwind.o): In function `find_proc_info':
>> > /home/jolsa/linux-perf/tools/perf/util/unwind.c:339: undefined reference to `_Ux86_64_dwarf_find_debug_frame'
>> > collect2: ld returned 1 exit status
>> > make: *** [perf] Error 1
>> >
>> >
>> > I'm using the latest code from git://git.sv.gnu.org/libunwind.git
>> >
>> > Looks like dwarf_find_debug_frame is not exported, although
>> > it looks like it is based on what I see in libunwind sources ;-)
>> >
>> > What did I miss?
>> libunwind needs to be configured with --enable-debug-frame for the
>> debug_frame code to be included in the lib.
>> On ARM the flag is always set while it isn't on x86. Here is the
>> culprit below (lines from libunwind configure).
>
> yay, thats it!
>
>>
>> Should that be changed in configure along with the changes in tools/perf?
>
> I guess it's ok
I wonder how to express the dependency between the perf changes and
the libunwind version and configure flags. Is a note in the commit
descritpion enough?
I fear that the change can break many build systems...
>
> Other than that tabs misformating the 'perf tool' change looks ok.
Ok I will respin the code with the typo and formatting changes.
>
> I tested the '.eh_frame' code and it's still working. Once
> I figure out how to create a x86 binary with .debug_frame
> data I'll test the code itself ;-)
Great!
>
> If you could think of any automated testcase for this that could
> be added under 'tests' that'd be nice (not necessarily)
Ok will look at it. Linaro has automated tests that I need to write as
well so rather do it in perf/tests directly ;-p
>
> thanks,
> jirka
Thanks!
Jean