scripts/decode_stacktrace.sh: Use LLVM environment variable

Message ID 20231215-llvm-decode-stacktrace-v1-1-201cb86f4879@quicinc.com
State New
Headers
Series scripts/decode_stacktrace.sh: Use LLVM environment variable |

Commit Message

Elliot Berman Dec. 15, 2023, 6:26 p.m. UTC
  When using LLVM as the compiler, decode_stacktrace should also use
llvm-addr2line. Check if LLVM is set and add the appropriate
suffix/prefix.

Signed-off-by: Elliot Berman <quic_eberman@quicinc.com>
---
 scripts/decode_stacktrace.sh | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)


---
base-commit: 3f7168591ebf7bbdb91797d02b1afaf00a4289b1
change-id: 20231214-llvm-decode-stacktrace-09538979006d

Best regards,
  

Comments

Justin Stitt Dec. 15, 2023, 9:19 p.m. UTC | #1
Hi,

On Fri, Dec 15, 2023 at 10:27 AM Elliot Berman <quic_eberman@quicinc.com> wrote:
>
> When using LLVM as the compiler, decode_stacktrace should also use
> llvm-addr2line. Check if LLVM is set and add the appropriate
> suffix/prefix.
>
> Signed-off-by: Elliot Berman <quic_eberman@quicinc.com>
> ---

I tried it out by dropping a random dump_stack() call into the init
phase of the kernel and redirecting the serial console output to a log
file. Using the pre-(this patch) and post-(this patch) script yields
good stack trace results. It seems llvm-addr2line is a drop-in
replacement.

Tested-by: Justin Stitt <justinstitt@google.com>

>  scripts/decode_stacktrace.sh | 12 +++++++++++-
>  1 file changed, 11 insertions(+), 1 deletion(-)
>
> diff --git a/scripts/decode_stacktrace.sh b/scripts/decode_stacktrace.sh
> index 564c5632e1a2..189b00f4e120 100755
> --- a/scripts/decode_stacktrace.sh
> +++ b/scripts/decode_stacktrace.sh
> @@ -16,6 +16,16 @@ elif type c++filt >/dev/null 2>&1 ; then
>         cppfilt_opts=-i
>  fi
>
> +if [[ "${LLVM}" == "1" ]] ; then
> +       addr2line="llvm-addr2line"
> +elif [[ "${LLVM}" == */ ]] ; then
> +       addr2line="${LLVM}llvm-addr2line"
> +elif [[ "${LLVM}" == -* ]] ; then
> +       addr2line="llvm-addr2line${LLVM}"
> +else
> +       addr2line="${CROSS_COMPILE}addr2line"
> +fi
> +
>  if [[ $1 == "-r" ]] ; then
>         vmlinux=""
>         basepath="auto"
> @@ -169,7 +179,7 @@ parse_symbol() {
>         if [[ $aarray_support == true && "${cache[$module,$address]+isset}" == "isset" ]]; then
>                 local code=${cache[$module,$address]}
>         else
> -               local code=$(${CROSS_COMPILE}addr2line -i -e "$objfile" "$address" 2>/dev/null)
> +               local code=$(${addr2line} -i -e "$objfile" "$address" 2>/dev/null)
>                 if [[ $aarray_support == true ]]; then
>                         cache[$module,$address]=$code
>                 fi
>
> ---
> base-commit: 3f7168591ebf7bbdb91797d02b1afaf00a4289b1
> change-id: 20231214-llvm-decode-stacktrace-09538979006d
>
> Best regards,
> --
> Elliot Berman <quic_eberman@quicinc.com>
>

Thanks
Justin
  
Nick Desaulniers Dec. 18, 2023, 4:37 p.m. UTC | #2
On Fri, Dec 15, 2023 at 10:27 AM Elliot Berman <quic_eberman@quicinc.com> wrote:

This patch looks familiar...
https://lore.kernel.org/lkml/20230929034836.403735-1-cmllamas@google.com/

Masahiro, can you please pick up that patch?

Elliot, Justin, can you help test/review that patch?

>
> When using LLVM as the compiler, decode_stacktrace should also use
> llvm-addr2line. Check if LLVM is set and add the appropriate
> suffix/prefix.
>
> Signed-off-by: Elliot Berman <quic_eberman@quicinc.com>
> ---
>  scripts/decode_stacktrace.sh | 12 +++++++++++-
>  1 file changed, 11 insertions(+), 1 deletion(-)
>
> diff --git a/scripts/decode_stacktrace.sh b/scripts/decode_stacktrace.sh
> index 564c5632e1a2..189b00f4e120 100755
> --- a/scripts/decode_stacktrace.sh
> +++ b/scripts/decode_stacktrace.sh
> @@ -16,6 +16,16 @@ elif type c++filt >/dev/null 2>&1 ; then
>         cppfilt_opts=-i
>  fi
>
> +if [[ "${LLVM}" == "1" ]] ; then
> +       addr2line="llvm-addr2line"
> +elif [[ "${LLVM}" == */ ]] ; then
> +       addr2line="${LLVM}llvm-addr2line"
> +elif [[ "${LLVM}" == -* ]] ; then
> +       addr2line="llvm-addr2line${LLVM}"
> +else
> +       addr2line="${CROSS_COMPILE}addr2line"
> +fi
> +
>  if [[ $1 == "-r" ]] ; then
>         vmlinux=""
>         basepath="auto"
> @@ -169,7 +179,7 @@ parse_symbol() {
>         if [[ $aarray_support == true && "${cache[$module,$address]+isset}" == "isset" ]]; then
>                 local code=${cache[$module,$address]}
>         else
> -               local code=$(${CROSS_COMPILE}addr2line -i -e "$objfile" "$address" 2>/dev/null)
> +               local code=$(${addr2line} -i -e "$objfile" "$address" 2>/dev/null)
>                 if [[ $aarray_support == true ]]; then
>                         cache[$module,$address]=$code
>                 fi
>
> ---
> base-commit: 3f7168591ebf7bbdb91797d02b1afaf00a4289b1
> change-id: 20231214-llvm-decode-stacktrace-09538979006d
>
> Best regards,
> --
> Elliot Berman <quic_eberman@quicinc.com>
>
  
Justin Stitt Dec. 19, 2023, 12:16 a.m. UTC | #3
Hi,

On Mon, Dec 18, 2023 at 8:37 AM Nick Desaulniers
<ndesaulniers@google.com> wrote:
>
> On Fri, Dec 15, 2023 at 10:27 AM Elliot Berman <quic_eberman@quicinc.com> wrote:
>
> This patch looks familiar...
> https://lore.kernel.org/lkml/20230929034836.403735-1-cmllamas@google.com/
>
> Masahiro, can you please pick up that patch?
>
> Elliot, Justin, can you help test/review that patch?
>

Interestingly, I am getting good stack traces on mainline with a
LLVM-built kernel -- both with and without that patch.

It may help others, so I'll go to that patch and add my tested-by.

> >
> > When using LLVM as the compiler, decode_stacktrace should also use
> > llvm-addr2line. Check if LLVM is set and add the appropriate
> > suffix/prefix.
> >
> > Signed-off-by: Elliot Berman <quic_eberman@quicinc.com>
> > ---
> >  scripts/decode_stacktrace.sh | 12 +++++++++++-
> >  1 file changed, 11 insertions(+), 1 deletion(-)
> >
> > diff --git a/scripts/decode_stacktrace.sh b/scripts/decode_stacktrace.sh
> > index 564c5632e1a2..189b00f4e120 100755
> > --- a/scripts/decode_stacktrace.sh
> > +++ b/scripts/decode_stacktrace.sh
> > @@ -16,6 +16,16 @@ elif type c++filt >/dev/null 2>&1 ; then
> >         cppfilt_opts=-i
> >  fi
> >
> > +if [[ "${LLVM}" == "1" ]] ; then
> > +       addr2line="llvm-addr2line"
> > +elif [[ "${LLVM}" == */ ]] ; then
> > +       addr2line="${LLVM}llvm-addr2line"
> > +elif [[ "${LLVM}" == -* ]] ; then
> > +       addr2line="llvm-addr2line${LLVM}"
> > +else
> > +       addr2line="${CROSS_COMPILE}addr2line"
> > +fi
> > +
> >  if [[ $1 == "-r" ]] ; then
> >         vmlinux=""
> >         basepath="auto"
> > @@ -169,7 +179,7 @@ parse_symbol() {
> >         if [[ $aarray_support == true && "${cache[$module,$address]+isset}" == "isset" ]]; then
> >                 local code=${cache[$module,$address]}
> >         else
> > -               local code=$(${CROSS_COMPILE}addr2line -i -e "$objfile" "$address" 2>/dev/null)
> > +               local code=$(${addr2line} -i -e "$objfile" "$address" 2>/dev/null)
> >                 if [[ $aarray_support == true ]]; then
> >                         cache[$module,$address]=$code
> >                 fi
> >
> > ---
> > base-commit: 3f7168591ebf7bbdb91797d02b1afaf00a4289b1
> > change-id: 20231214-llvm-decode-stacktrace-09538979006d
> >
> > Best regards,
> > --
> > Elliot Berman <quic_eberman@quicinc.com>
> >
>
>
> --
> Thanks,
> ~Nick Desaulniers

Thanks
Justin
  
Nick Desaulniers Dec. 19, 2023, 4:49 p.m. UTC | #4
On Mon, Dec 18, 2023 at 4:17 PM Justin Stitt <justinstitt@google.com> wrote:
>
> Interestingly, I am getting good stack traces on mainline with a
> LLVM-built kernel -- both with and without that patch.

Probably because:
1. you have GNU binutils installed.
2. you're not testing .o files from an LTO build (which GNU binutils
cannot decode).
  
Justin Stitt Dec. 19, 2023, 9:12 p.m. UTC | #5
On Tue, Dec 19, 2023 at 8:49 AM Nick Desaulniers
<ndesaulniers@google.com> wrote:
>
> On Mon, Dec 18, 2023 at 4:17 PM Justin Stitt <justinstitt@google.com> wrote:
> >
> > Interestingly, I am getting good stack traces on mainline with a
> > LLVM-built kernel -- both with and without that patch.
>
> Probably because:
> 1. you have GNU binutils installed.
> 2. you're not testing .o files from an LTO build (which GNU binutils
> cannot decode).
>

Yes, and yes. Makes sense.

> --
> Thanks,
> ~Nick Desaulniers

Cheers
  

Patch

diff --git a/scripts/decode_stacktrace.sh b/scripts/decode_stacktrace.sh
index 564c5632e1a2..189b00f4e120 100755
--- a/scripts/decode_stacktrace.sh
+++ b/scripts/decode_stacktrace.sh
@@ -16,6 +16,16 @@  elif type c++filt >/dev/null 2>&1 ; then
 	cppfilt_opts=-i
 fi
 
+if [[ "${LLVM}" == "1" ]] ; then
+	addr2line="llvm-addr2line"
+elif [[ "${LLVM}" == */ ]] ; then
+	addr2line="${LLVM}llvm-addr2line"
+elif [[ "${LLVM}" == -* ]] ; then
+	addr2line="llvm-addr2line${LLVM}"
+else
+	addr2line="${CROSS_COMPILE}addr2line"
+fi
+
 if [[ $1 == "-r" ]] ; then
 	vmlinux=""
 	basepath="auto"
@@ -169,7 +179,7 @@  parse_symbol() {
 	if [[ $aarray_support == true && "${cache[$module,$address]+isset}" == "isset" ]]; then
 		local code=${cache[$module,$address]}
 	else
-		local code=$(${CROSS_COMPILE}addr2line -i -e "$objfile" "$address" 2>/dev/null)
+		local code=$(${addr2line} -i -e "$objfile" "$address" 2>/dev/null)
 		if [[ $aarray_support == true ]]; then
 			cache[$module,$address]=$code
 		fi