x86_64-m32 internal error for multi-thread-step.exp [Re: [PATCH v10 06/28] btrace: change branch trace data structure]

[Jan Kratochvil <jan dot kratochvil at redhat dot com>]

On Tue, 14 Jan 2014 09:04:16 +0100, Markus Metzger wrote:
[...]
> +/* Allocate and initialize a new branch trace function segment.
> +   PREV is the chronologically preceding function segment.
> +   MFUN and FUN are the symbol information we have for this function.  */
> +
> +static struct btrace_function *
> +ftrace_new_function (struct btrace_function *prev,
> +		     struct minimal_symbol *mfun,
> +		     struct symbol *fun)
> +{
> +  struct btrace_function *bfun;
> +
> +  bfun = xzalloc (sizeof (*bfun));
> +
> +  bfun->msym = mfun;
> +  bfun->sym = fun;
> +  bfun->flow.prev = prev;
> +
> +  /* We start with the identities of min and max, respectively.  */
> +  bfun->lbegin = INT_MAX;
> +  bfun->lend = INT_MIN;
>  
> -  if (filename == NULL)
> -    filename = "";
> +  if (prev != NULL)
> +    {
> +      gdb_assert (prev->flow.next == NULL);

(gdb) PASS: gdb.btrace/multi-thread-step.exp: navigate: thread 1: record goto begin
info record^M
Active record target: record-btrace^M
btrace.c:220: internal-error: ftrace_new_function: Assertion `prev->flow.next == NULL' failed.^M

This occasionally happens on x86_64 system with -m32 inferior:
	runtest CC_FOR_TARGET="gcc -m32" CXX_FOR_TARGET="g++ -m32" gdb.btrace/multi-thread-step.exp

cpu family	: 6
model		: 63
model name	: Intel(R) Xeon(R) CPU E5-2630 v3 @ 2.40GHz
stepping	: 2
microcode	: 0x2a


> +      prev->flow.next = bfun;
>  
> -  return (filename_cmp (bfile, filename) != 0);
> +      bfun->number = prev->number + 1;
> +      bfun->insn_offset = (prev->insn_offset
> +			   + VEC_length (btrace_insn_s, prev->insn));
> +    }
> +
> +  return bfun;
>  }

Although for -m32 inferior there are 100% reproducible more FAILs:
	FAIL: gdb.btrace/rn-dl-bind.exp: reverse-next
	FAIL: gdb.btrace/rn-dl-bind.exp: reverse-step
	FAIL: gdb.btrace/rn-dl-bind.exp: reverse-step
	FAIL: gdb.btrace/rn-dl-bind.exp: reverse-next
	FAIL: gdb.btrace/rn-dl-bind.exp: next (the program exited)
	FAIL: gdb.btrace/exception.exp: flat (pattern 1)
	FAIL: gdb.btrace/exception.exp: indented (pattern 1) 


So I do not think the assert may be such a critical issue as IIUC btrace is
probably not used for 32-bit inferiors.


Regards,
Jan