How to disable GDB backtrace new behavior of including disassembly in backtrace ?

How to disable GDB backtrace new behavior of including disassembly in backtrace ?

[William Tambe]

How to disable GDB backtrace new behavior of including disassembly in
backtrace ?

I have built GDB 8.3 and here is an example of output that I now see:

bt
#0  0x0005ed42 in set_next_entity (se=<optimized out>,
cfs_rq=<optimized out>) at /linux/kernel/sched/fair.c:6844
   0x0005ed38 <pick_next_task_fair+492>:        80 74   li8 %7, 0x4 # 4
   0x0005ed3a <pick_next_task_fair+494>:        b8 70   add %7, %sp
   0x0005ed3c <pick_next_task_fair+496>:        ea 37   ld32 %3, %7
   0x0005ed3e <pick_next_task_fair+498>:        c7 13   cpy %1, %3
   0x0005ed40 <pick_next_task_fair+500>:        92 10   inc8 %1, 32 # 0x20
=> 0x0005ed42 <pick_next_task_fair+502>:        ea 11   ld32 %1, %1
   0x0005ed44 <pick_next_task_fair+504>:        ea 50   ld32 %5, %sp
   0x0005ed46 <pick_next_task_fair+506>:        e0 d0   gip %sr
   0x0005ed48 <pick_next_task_fair+508>:        a2 d0 4e 00 00 00
 inc32 %sr, 78 # 0x4e
   0x0005ed4e <pick_next_task_fair+514>:        d1 1d   jnz %1, %sr
#1  pick_next_task_fair (rq=<optimized out>, prev=<optimized out>,
rf=<optimized out>) at /linux/kernel/sched/fair.c:6844

Using an earlier version of GDB would have just shown two lines
without disassembly; ie:

bt
#0  0x0005ed42 in set_next_entity (se=<optimized out>,
cfs_rq=<optimized out>) at /linux/kernel/sched/fair.c:6844
#1  pick_next_task_fair (rq=<optimized out>, prev=<optimized out>,
rf=<optimized out>) at /linux/kernel/sched/fair.c:6844

How can I restore the GDB old behavior when using backtrace ?

Re: How to disable GDB backtrace new behavior of including disassembly in backtrace ?

[Simon Marchi]

On 2019-12-13 7:09 a.m., William Tambe wrote:
> How to disable GDB backtrace new behavior of including disassembly in
> backtrace ?
> 
> I have built GDB 8.3 and here is an example of output that I now see:
> 
> bt
> #0  0x0005ed42 in set_next_entity (se=<optimized out>,
> cfs_rq=<optimized out>) at /linux/kernel/sched/fair.c:6844
>    0x0005ed38 <pick_next_task_fair+492>:        80 74   li8 %7, 0x4 # 4
>    0x0005ed3a <pick_next_task_fair+494>:        b8 70   add %7, %sp
>    0x0005ed3c <pick_next_task_fair+496>:        ea 37   ld32 %3, %7
>    0x0005ed3e <pick_next_task_fair+498>:        c7 13   cpy %1, %3
>    0x0005ed40 <pick_next_task_fair+500>:        92 10   inc8 %1, 32 # 0x20
> => 0x0005ed42 <pick_next_task_fair+502>:        ea 11   ld32 %1, %1
>    0x0005ed44 <pick_next_task_fair+504>:        ea 50   ld32 %5, %sp
>    0x0005ed46 <pick_next_task_fair+506>:        e0 d0   gip %sr
>    0x0005ed48 <pick_next_task_fair+508>:        a2 d0 4e 00 00 00
>  inc32 %sr, 78 # 0x4e
>    0x0005ed4e <pick_next_task_fair+514>:        d1 1d   jnz %1, %sr
> #1  pick_next_task_fair (rq=<optimized out>, prev=<optimized out>,
> rf=<optimized out>) at /linux/kernel/sched/fair.c:6844
> 
> Using an earlier version of GDB would have just shown two lines
> without disassembly; ie:
> 
> bt
> #0  0x0005ed42 in set_next_entity (se=<optimized out>,
> cfs_rq=<optimized out>) at /linux/kernel/sched/fair.c:6844
> #1  pick_next_task_fair (rq=<optimized out>, prev=<optimized out>,
> rf=<optimized out>) at /linux/kernel/sched/fair.c:6844
> 
> How can I restore the GDB old behavior when using backtrace ?
> 

Hi William,

I have never seen this behavior.  It's either a setting I have never
encountered, or a Python script you have loaded that is providing this
functionality (e.g. with a frame-filter).

Are you able to reproduce this behavior with another program?  If so, could
you send a small reproducer (small source snippet to compile + GDB commands),
we can try it and tell you if we see the same behavior.

Simon

Re: How to disable GDB backtrace new behavior of including disassembly in backtrace ?

[William Tambe]

On Fri, Dec 13, 2019 at 2:48 PM Simon Marchi <simark@simark.ca> wrote:
>
> On 2019-12-13 7:09 a.m., William Tambe wrote:
> > How to disable GDB backtrace new behavior of including disassembly in
> > backtrace ?
> >
> > I have built GDB 8.3 and here is an example of output that I now see:
> >
> > bt
> > #0  0x0005ed42 in set_next_entity (se=<optimized out>,
> > cfs_rq=<optimized out>) at /linux/kernel/sched/fair.c:6844
> >    0x0005ed38 <pick_next_task_fair+492>:        80 74   li8 %7, 0x4 # 4
> >    0x0005ed3a <pick_next_task_fair+494>:        b8 70   add %7, %sp
> >    0x0005ed3c <pick_next_task_fair+496>:        ea 37   ld32 %3, %7
> >    0x0005ed3e <pick_next_task_fair+498>:        c7 13   cpy %1, %3
> >    0x0005ed40 <pick_next_task_fair+500>:        92 10   inc8 %1, 32 # 0x20
> > => 0x0005ed42 <pick_next_task_fair+502>:        ea 11   ld32 %1, %1
> >    0x0005ed44 <pick_next_task_fair+504>:        ea 50   ld32 %5, %sp
> >    0x0005ed46 <pick_next_task_fair+506>:        e0 d0   gip %sr
> >    0x0005ed48 <pick_next_task_fair+508>:        a2 d0 4e 00 00 00
> >  inc32 %sr, 78 # 0x4e
> >    0x0005ed4e <pick_next_task_fair+514>:        d1 1d   jnz %1, %sr
> > #1  pick_next_task_fair (rq=<optimized out>, prev=<optimized out>,
> > rf=<optimized out>) at /linux/kernel/sched/fair.c:6844
> >
> > Using an earlier version of GDB would have just shown two lines
> > without disassembly; ie:
> >
> > bt
> > #0  0x0005ed42 in set_next_entity (se=<optimized out>,
> > cfs_rq=<optimized out>) at /linux/kernel/sched/fair.c:6844
> > #1  pick_next_task_fair (rq=<optimized out>, prev=<optimized out>,
> > rf=<optimized out>) at /linux/kernel/sched/fair.c:6844
> >
> > How can I restore the GDB old behavior when using backtrace ?
> >
>
> Hi William,
>
> I have never seen this behavior.  It's either a setting I have never
> encountered, or a Python script you have loaded that is providing this
> functionality (e.g. with a frame-filter).

I do not have a python script loaded.
What GDB command should I use to check whether I have a frame-filter ?

>
> Are you able to reproduce this behavior with another program?  If so, could
> you send a small reproducer (small source snippet to compile + GDB commands),

Yes, I am able to reproduce it with any program that I compile; here
is a simple example where I can see the issue when breaking after
calling printf() and using the command backtrace.
---
#include <stdio.h>
void myfunc() {
        printf("Hello world!\n");
}
void main () {
        myfunc();
}
---

> we can try it and tell you if we see the same behavior.

I am using the following Binutils release that include GDB:

commit b5624945ea67525c0ba4ffec7a9d3f9366bf9071 (tag: binutils-2_33_1)
Author: Nick Clifton <nickc@redhat.com>
Date:   Sat Oct 12 15:07:18 2019 +0100

    Release 2.33.1

>
> Simon

Re: How to disable GDB backtrace new behavior of including disassembly in backtrace ?

[Simon Marchi]

On 2019-12-24 7:30 a.m., William Tambe wrote:
> I do not have a python script loaded.
> What GDB command should I use to check whether I have a frame-filter ?

"info frame-filter"

You can also try to start gdb with `-nx`, which will inhibit it from reading
the various gdbinit files that could change the settings.  If that makes a
difference, it means that some gdbinit file (either in your home or system wide)
is responsible.

> 
>>
>> Are you able to reproduce this behavior with another program?  If so, could
>> you send a small reproducer (small source snippet to compile + GDB commands),
> 
> Yes, I am able to reproduce it with any program that I compile; here
> is a simple example where I can see the issue when breaking after
> calling printf() and using the command backtrace.
> ---
> #include <stdio.h>
> void myfunc() {
>         printf("Hello world!\n");
> }
> void main () {
>         myfunc();
> }

This is what I get:

Reading symbols from a.out...
(gdb) start
Temporary breakpoint 1 at 0x1150: file allo.c, line 6.
Starting program: /home/simark/build/binutils-gdb/gdb/a.out

Temporary breakpoint 1, main () at allo.c:6
6	        myfunc();
(gdb) s
myfunc () at allo.c:3
3	        printf("Hello world!\n");
(gdb) bt
#0  myfunc () at allo.c:3
#1  0x000055555555515a in main () at allo.c:6

Simon