2024-05-21 00:57:45

by Xiong Nandi

[permalink] [raw]
Subject: [PATCH] scripts/decode_stacktrace.sh: better support to ARM32 module stack trace

Since System.map is generated by cross-compile nm tool, we should use it
here too. Otherwise host nm may not recognize thumb2 function address well.

Beside, sometimes special characters around module name, such as ARM32
with BACKTRACE_VERBOSE in "(%pS)" format, such as:
[<806e4845>] (dump_stack_lvl) from [<7f806013>] (hello_init+0x13/0x1000 [test])

After stripping other characters around "[module]", it can be finally decoded:
(dump_stack_lvl) from hello_init (/foo/test.c:10) test

Signed-off-by: xndcn <[email protected]>
---
scripts/decode_stacktrace.sh | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/scripts/decode_stacktrace.sh b/scripts/decode_stacktrace.sh
index fa5be6f57b0..324e4a6c260 100755
--- a/scripts/decode_stacktrace.sh
+++ b/scripts/decode_stacktrace.sh
@@ -30,6 +30,7 @@ fi

READELF=${UTIL_PREFIX}readelf${UTIL_SUFFIX}
ADDR2LINE=${UTIL_PREFIX}addr2line${UTIL_SUFFIX}
+NM=${UTIL_PREFIX}nm${UTIL_SUFFIX}

if [[ $1 == "-r" ]] ; then
vmlinux=""
@@ -158,7 +159,7 @@ parse_symbol() {
if [[ $aarray_support == true && "${cache[$module,$name]+isset}" == "isset" ]]; then
local base_addr=${cache[$module,$name]}
else
- local base_addr=$(nm "$objfile" 2>/dev/null | awk '$3 == "'$name'" && ($2 == "t" || $2 == "T") {print $1; exit}')
+ local base_addr=$(${NM} "$objfile" 2>/dev/null | awk '$3 == "'$name'" && ($2 == "t" || $2 == "T") {print $1; exit}')
if [[ $base_addr == "" ]] ; then
# address not found
return
@@ -282,8 +283,8 @@ handle_line() {

if [[ ${words[$last]} =~ \[([^]]+)\] ]]; then
module=${words[$last]}
- module=${module#\[}
- module=${module%\]}
+ module=${module#*\[}
+ module=${module%\]*}
modbuildid=${module#* }
module=${module% *}
if [[ $modbuildid == $module ]]; then
--
2.25.1



2024-05-22 01:27:51

by Xiong Nandi

[permalink] [raw]
Subject: Re: [PATCH] scripts/decode_stacktrace.sh: better support to ARM32 module stack trace

Sorry about the name, it is some kind of abbreviation. So I re-post here:
---
Since System.map is generated by cross-compile nm tool, we should use it
here too. Otherwise host nm may not recognize thumb2 function address well.

Beside, sometimes special characters around module name, such as ARM32
with BACKTRACE_VERBOSE in "(%pS)" format, such as:
[<806e4845>] (dump_stack_lvl) from [<7f806013>] (hello_init+0x13/0x1000 [test])

After stripping other characters around "[module]", it can be finally decoded:
(dump_stack_lvl) from hello_init (/foo/test.c:10) test

Signed-off-by: Xiong Nandi <[email protected]>
---
scripts/decode_stacktrace.sh | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/scripts/decode_stacktrace.sh b/scripts/decode_stacktrace.sh
index fa5be6f57b00..324e4a6c260a 100755
--- a/scripts/decode_stacktrace.sh
+++ b/scripts/decode_stacktrace.sh
@@ -30,6 +30,7 @@ fi

READELF=${UTIL_PREFIX}readelf${UTIL_SUFFIX}
ADDR2LINE=${UTIL_PREFIX}addr2line${UTIL_SUFFIX}
+NM=${UTIL_PREFIX}nm${UTIL_SUFFIX}

if [[ $1 == "-r" ]] ; then
vmlinux=""
@@ -158,7 +159,7 @@ parse_symbol() {
if [[ $aarray_support == true && "${cache[$module,$name]+isset}" == "isset" ]]; then
local base_addr=${cache[$module,$name]}
else
- local base_addr=$(nm "$objfile" 2>/dev/null | awk '$3 == "'$name'" && ($2 == "t" || $2 == "T") {print $1; exit}')
+ local base_addr=$(${NM} "$objfile" 2>/dev/null | awk '$3 == "'$name'" && ($2 == "t" || $2 == "T") {print $1; exit}')
if [[ $base_addr == "" ]] ; then
# address not found
return
@@ -282,8 +283,8 @@ handle_line() {

if [[ ${words[$last]} =~ \[([^]]+)\] ]]; then
module=${words[$last]}
- module=${module#\[}
- module=${module%\]}
+ module=${module#*\[}
+ module=${module%\]*}
modbuildid=${module#* }
module=${module% *}
if [[ $modbuildid == $module ]]; then
--
2.25.1


2024-05-22 02:49:17

by Elliot Berman

[permalink] [raw]
Subject: Re: [PATCH] scripts/decode_stacktrace.sh: better support to ARM32 module stack trace

On Wed, May 22, 2024 at 09:05:59AM +0800, Xiong Nandi wrote:
> Sorry about the name, it is some kind of abbreviation. So I re-post here:
> ---
> Since System.map is generated by cross-compile nm tool, we should use it
> here too. Otherwise host nm may not recognize thumb2 function address well.
>
> Beside, sometimes special characters around module name, such as ARM32
> with BACKTRACE_VERBOSE in "(%pS)" format, such as:
> [<806e4845>] (dump_stack_lvl) from [<7f806013>] (hello_init+0x13/0x1000 [test])
>
> After stripping other characters around "[module]", it can be finally decoded:
> (dump_stack_lvl) from hello_init (/foo/test.c:10) test
>
> Signed-off-by: Xiong Nandi <[email protected]>
> ---
> scripts/decode_stacktrace.sh | 7 ++++---
> 1 file changed, 4 insertions(+), 3 deletions(-)
>
> diff --git a/scripts/decode_stacktrace.sh b/scripts/decode_stacktrace.sh
> index fa5be6f57b00..324e4a6c260a 100755
> --- a/scripts/decode_stacktrace.sh
> +++ b/scripts/decode_stacktrace.sh
> @@ -30,6 +30,7 @@ fi
>
> READELF=${UTIL_PREFIX}readelf${UTIL_SUFFIX}
> ADDR2LINE=${UTIL_PREFIX}addr2line${UTIL_SUFFIX}
> +NM=${UTIL_PREFIX}nm${UTIL_SUFFIX}
>
> if [[ $1 == "-r" ]] ; then
> vmlinux=""
> @@ -158,7 +159,7 @@ parse_symbol() {
> if [[ $aarray_support == true && "${cache[$module,$name]+isset}" == "isset" ]]; then
> local base_addr=${cache[$module,$name]}
> else
> - local base_addr=$(nm "$objfile" 2>/dev/null | awk '$3 == "'$name'" && ($2 == "t" || $2 == "T") {print $1; exit}')
> + local base_addr=$(${NM} "$objfile" 2>/dev/null | awk '$3 == "'$name'" && ($2 == "t" || $2 == "T") {print $1; exit}')

The nm parts should be a separate patch.

> if [[ $base_addr == "" ]] ; then
> # address not found
> return
> @@ -282,8 +283,8 @@ handle_line() {
>
> if [[ ${words[$last]} =~ \[([^]]+)\] ]]; then
> module=${words[$last]}
> - module=${module#\[}
> - module=${module%\]}
> + module=${module#*\[}
> + module=${module%\]*}

I need to get a moment to play with it. Is my understanding correct that
the problem is that the last word ($module) is:

[test])

and after the existing strip logic, $module becomes test]) whereas
expecting just "test"? Your change is to strip any leading/trailing
characters before/after the [ / ] respectively? Isn't this a problem for
$symbol as well -- it would be "(hello_init+0x13/0x1000" in the example.

- Elliot

> modbuildid=${module#* }
> module=${module% *}
> if [[ $modbuildid == $module ]]; then
> --
> 2.25.1
>

2024-05-22 04:10:48

by Xiong Nandi

[permalink] [raw]
Subject: Re: [PATCH] scripts/decode_stacktrace.sh: better support to ARM32 module stack trace

Thanks, I'll split it into 2 patches later.

> Your change is to strip any leading/trailing characters before/after the [ / ] respectively?
Yes, exactly.

> Isn't this a problem for $symbol as well
there is already a strip logic in "parse_symbol()", which seems
introduced by #e260fe01, also for ARM.
> # Remove the englobing parenthesis
> symbol=${symbol#\(}
> symbol=${symbol%\)}

On Wed, May 22, 2024 at 10:48 AM Elliot Berman <[email protected]> wrote:
>
> On Wed, May 22, 2024 at 09:05:59AM +0800, Xiong Nandi wrote:
> > Sorry about the name, it is some kind of abbreviation. So I re-post here:
> > ---
> > Since System.map is generated by cross-compile nm tool, we should use it
> > here too. Otherwise host nm may not recognize thumb2 function address well.
> >
> > Beside, sometimes special characters around module name, such as ARM32
> > with BACKTRACE_VERBOSE in "(%pS)" format, such as:
> > [<806e4845>] (dump_stack_lvl) from [<7f806013>] (hello_init+0x13/0x1000 [test])
> >
> > After stripping other characters around "[module]", it can be finally decoded:
> > (dump_stack_lvl) from hello_init (/foo/test.c:10) test
> >
> > Signed-off-by: Xiong Nandi <[email protected]>
> > ---
> > scripts/decode_stacktrace.sh | 7 ++++---
> > 1 file changed, 4 insertions(+), 3 deletions(-)
> >
> > diff --git a/scripts/decode_stacktrace.sh b/scripts/decode_stacktrace.sh
> > index fa5be6f57b00..324e4a6c260a 100755
> > --- a/scripts/decode_stacktrace.sh
> > +++ b/scripts/decode_stacktrace.sh
> > @@ -30,6 +30,7 @@ fi
> >
> > READELF=${UTIL_PREFIX}readelf${UTIL_SUFFIX}
> > ADDR2LINE=${UTIL_PREFIX}addr2line${UTIL_SUFFIX}
> > +NM=${UTIL_PREFIX}nm${UTIL_SUFFIX}
> >
> > if [[ $1 == "-r" ]] ; then
> > vmlinux=""
> > @@ -158,7 +159,7 @@ parse_symbol() {
> > if [[ $aarray_support == true && "${cache[$module,$name]+isset}" == "isset" ]]; then
> > local base_addr=${cache[$module,$name]}
> > else
> > - local base_addr=$(nm "$objfile" 2>/dev/null | awk '$3 == "'$name'" && ($2 == "t" || $2 == "T") {print $1; exit}')
> > + local base_addr=$(${NM} "$objfile" 2>/dev/null | awk '$3 == "'$name'" && ($2 == "t" || $2 == "T") {print $1; exit}')
>
> The nm parts should be a separate patch.
>
> > if [[ $base_addr == "" ]] ; then
> > # address not found
> > return
> > @@ -282,8 +283,8 @@ handle_line() {
> >
> > if [[ ${words[$last]} =~ \[([^]]+)\] ]]; then
> > module=${words[$last]}
> > - module=${module#\[}
> > - module=${module%\]}
> > + module=${module#*\[}
> > + module=${module%\]*}
>
> I need to get a moment to play with it. Is my understanding correct that
> the problem is that the last word ($module) is:
>
> [test])
>
> and after the existing strip logic, $module becomes test]) whereas
> expecting just "test"? Your change is to strip any leading/trailing
> characters before/after the [ / ] respectively? Isn't this a problem for
> $symbol as well -- it would be "(hello_init+0x13/0x1000" in the example.
>
> - Elliot
>
> > modbuildid=${module#* }
> > module=${module% *}
> > if [[ $modbuildid == $module ]]; then
> > --
> > 2.25.1
> >

2024-05-23 01:03:44

by Xiong Nandi

[permalink] [raw]
Subject: [PATCH v2 0/2] scripts/decode_stacktrace.sh: better support to ARM32

v2:
- Split the patch into two.

Xiong Nandi (2):
scripts/decode_stacktrace.sh: better support to ARM32 module stack trace
scripts/decode_stacktrace.sh: wrap nm with UTIL_PREFIX and UTIL_SUFFIX

scripts/decode_stacktrace.sh | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)
--
2.25.1


2024-05-23 01:03:55

by Xiong Nandi

[permalink] [raw]
Subject: [PATCH v2 1/2] scripts/decode_stacktrace.sh: wrap nm with UTIL_PREFIX and UTIL_SUFFIX

Since System.map is generated by cross-compile nm tool, we should use it here
too. Otherwise host nm may not recognize ARM Thumb-2 instruction address well.

Signed-off-by: Xiong Nandi <[email protected]>
---
scripts/decode_stacktrace.sh | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/scripts/decode_stacktrace.sh b/scripts/decode_stacktrace.sh
index fa5be6f57b00..2bc3a54ffba5 100755
--- a/scripts/decode_stacktrace.sh
+++ b/scripts/decode_stacktrace.sh
@@ -30,6 +30,7 @@ fi

READELF=${UTIL_PREFIX}readelf${UTIL_SUFFIX}
ADDR2LINE=${UTIL_PREFIX}addr2line${UTIL_SUFFIX}
+NM=${UTIL_PREFIX}nm${UTIL_SUFFIX}

if [[ $1 == "-r" ]] ; then
vmlinux=""
@@ -158,7 +159,7 @@ parse_symbol() {
if [[ $aarray_support == true && "${cache[$module,$name]+isset}" == "isset" ]]; then
local base_addr=${cache[$module,$name]}
else
- local base_addr=$(nm "$objfile" 2>/dev/null | awk '$3 == "'$name'" && ($2 == "t" || $2 == "T") {print $1; exit}')
+ local base_addr=$(${NM} "$objfile" 2>/dev/null | awk '$3 == "'$name'" && ($2 == "t" || $2 == "T") {print $1; exit}')
if [[ $base_addr == "" ]] ; then
# address not found
return
--
2.25.1


2024-05-23 01:04:05

by Xiong Nandi

[permalink] [raw]
Subject: [PATCH v2 2/2] scripts/decode_stacktrace.sh: better support to ARM32 module stack trace

Sometimes there is special characters around module name in stack trace,
such as ARM32 with BACKTRACE_VERBOSE in "(%pS)" format, such as:
[<806e4845>] (dump_stack_lvl) from [<7f806013>] (hello_init+0x13/0x1000 [test])

After stripping other characters around "[module]", it can be finally decoded:
(dump_stack_lvl) from hello_init (/foo/test.c:10) test

Signed-off-by: Xiong Nandi <[email protected]>
---
scripts/decode_stacktrace.sh | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/scripts/decode_stacktrace.sh b/scripts/decode_stacktrace.sh
index 2bc3a54ffba5..324e4a6c260a 100755
--- a/scripts/decode_stacktrace.sh
+++ b/scripts/decode_stacktrace.sh
@@ -283,8 +283,8 @@ handle_line() {

if [[ ${words[$last]} =~ \[([^]]+)\] ]]; then
module=${words[$last]}
- module=${module#\[}
- module=${module%\]}
+ module=${module#*\[}
+ module=${module%\]*}
modbuildid=${module#* }
module=${module% *}
if [[ $modbuildid == $module ]]; then
--
2.25.1


2024-05-23 01:56:49

by Elliot Berman

[permalink] [raw]
Subject: Re: [PATCH v2 1/2] scripts/decode_stacktrace.sh: wrap nm with UTIL_PREFIX and UTIL_SUFFIX

On Thu, May 23, 2024 at 09:03:17AM +0800, Xiong Nandi wrote:
> Since System.map is generated by cross-compile nm tool, we should use it here
> too. Otherwise host nm may not recognize ARM Thumb-2 instruction address well.
>
> Signed-off-by: Xiong Nandi <[email protected]>

Reviewed-by: Elliot Berman <[email protected]>

> ---
> scripts/decode_stacktrace.sh | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/scripts/decode_stacktrace.sh b/scripts/decode_stacktrace.sh
> index fa5be6f57b00..2bc3a54ffba5 100755
> --- a/scripts/decode_stacktrace.sh
> +++ b/scripts/decode_stacktrace.sh
> @@ -30,6 +30,7 @@ fi
>
> READELF=${UTIL_PREFIX}readelf${UTIL_SUFFIX}
> ADDR2LINE=${UTIL_PREFIX}addr2line${UTIL_SUFFIX}
> +NM=${UTIL_PREFIX}nm${UTIL_SUFFIX}
>
> if [[ $1 == "-r" ]] ; then
> vmlinux=""
> @@ -158,7 +159,7 @@ parse_symbol() {
> if [[ $aarray_support == true && "${cache[$module,$name]+isset}" == "isset" ]]; then
> local base_addr=${cache[$module,$name]}
> else
> - local base_addr=$(nm "$objfile" 2>/dev/null | awk '$3 == "'$name'" && ($2 == "t" || $2 == "T") {print $1; exit}')
> + local base_addr=$(${NM} "$objfile" 2>/dev/null | awk '$3 == "'$name'" && ($2 == "t" || $2 == "T") {print $1; exit}')
> if [[ $base_addr == "" ]] ; then
> # address not found
> return
> --
> 2.25.1
>

2024-05-23 02:00:47

by Elliot Berman

[permalink] [raw]
Subject: Re: [PATCH v2 2/2] scripts/decode_stacktrace.sh: better support to ARM32 module stack trace

On Thu, May 23, 2024 at 09:03:18AM +0800, Xiong Nandi wrote:
> Sometimes there is special characters around module name in stack trace,
> such as ARM32 with BACKTRACE_VERBOSE in "(%pS)" format, such as:
> [<806e4845>] (dump_stack_lvl) from [<7f806013>] (hello_init+0x13/0x1000 [test])
>
> After stripping other characters around "[module]", it can be finally decoded:
> (dump_stack_lvl) from hello_init (/foo/test.c:10) test
>
> Signed-off-by: Xiong Nandi <[email protected]>
> ---
> scripts/decode_stacktrace.sh | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/scripts/decode_stacktrace.sh b/scripts/decode_stacktrace.sh
> index 2bc3a54ffba5..324e4a6c260a 100755
> --- a/scripts/decode_stacktrace.sh
> +++ b/scripts/decode_stacktrace.sh
> @@ -283,8 +283,8 @@ handle_line() {
>
> if [[ ${words[$last]} =~ \[([^]]+)\] ]]; then
> module=${words[$last]}
> - module=${module#\[}
> - module=${module%\]}
> + module=${module#*\[}
> + module=${module%\]*}

I think it'd be better to just remove the parenthesis similar to how is
done in the symbol name.

That is:

module=${words[$last]}
module=${module#\[}
module=${module%\]}
# some nice comment explaining only the closing paren is
# need to be stripped
module=${module%\)}
modbuildid=${module#* }

> modbuildid=${module#* }
> module=${module% *}
> if [[ $modbuildid == $module ]]; then
> --
> 2.25.1
>