Re: [PATCH v2] gdb/arm: Stop unwinding on error, but do not assert

[Luis Machado <luis.machado@arm.com>]

Hi,

On 10/13/22 10:17, Torbjörn SVENSSON wrote:
> When it's impossible to read the FPCCR and XPSR, the unwinding is
> unpredictable as the it's not possible to determine the correct
> frame size or padding.
> The only sane thing to do in this condition is to stop the unwinding.
> 
> Without this patch, gdb would assert if this errornous state was
> detected.
> 

Could you please attach an example of the change in the commit message? What it does before
and after the change?

> Signed-off-by: Torbjörn SVENSSON  <torbjorn.svensson@foss.st.com>
> ---
>   gdb/arm-tdep.c | 35 +++++++++++++++++++++++++++++------
>   1 file changed, 29 insertions(+), 6 deletions(-)
> 
> diff --git a/gdb/arm-tdep.c b/gdb/arm-tdep.c
> index 041e6afefed..afcbce478c2 100644
> --- a/gdb/arm-tdep.c
> +++ b/gdb/arm-tdep.c
> @@ -3591,9 +3591,13 @@ arm_m_exception_cache (frame_info_ptr this_frame)
>   	  ULONGEST fpcar;
>   
>   	  /* Read FPCCR register.  */
> -	  gdb_assert (safe_read_memory_unsigned_integer (FPCCR,
> -							 ARM_INT_REGISTER_SIZE,
> -							 byte_order, &fpccr));
> +	 if (!safe_read_memory_unsigned_integer (FPCCR, ARM_INT_REGISTER_SIZE,
> +						 byte_order, &fpccr))
> +	   {
> +	     warning (_("Could not fetch required FPCCR content.  Further "
> +			"unwind is impossible."));
> +	     return NULL;


NULL -> nullptr everywhere.

> +	   }
>   
>   	  /* Read FPCAR register.  */
>   	  if (!safe_read_memory_unsigned_integer (FPCAR, ARM_INT_REGISTER_SIZE,
> @@ -3669,9 +3673,15 @@ arm_m_exception_cache (frame_info_ptr this_frame)
>   	 aligner between the top of the 32-byte stack frame and the
>   	 previous context's stack pointer.  */
>         ULONGEST xpsr;
> -      gdb_assert (safe_read_memory_unsigned_integer (cache->saved_regs[
> -						     ARM_PS_REGNUM].addr (), 4,
> -						     byte_order, &xpsr));
> +      if (!safe_read_memory_unsigned_integer (cache->saved_regs[ARM_PS_REGNUM]
> +					      .addr (), ARM_INT_REGISTER_SIZE,
> +					      byte_order, &xpsr))
> +	{
> +	  warning (_("Could not fetch required XPSR content.  Further unwind "
> +		     "is impossible."));
> +	  return NULL;
> +	}
> +
>         if (bit (xpsr, 9) != 0)
>   	{
>   	  CORE_ADDR new_sp = arm_cache_get_prev_sp_value (cache, tdep) + 4;
> @@ -3703,6 +3713,14 @@ arm_m_exception_this_id (frame_info_ptr this_frame,
>       *this_cache = arm_m_exception_cache (this_frame);
>     cache = (struct arm_prologue_cache *) *this_cache;
>   
> +  /* Unwind of this frame is not possible.  Return outer_frame_id to stop the
> +     unwinding.  */
> +  if (cache == NULL)
> +    {
> +      *this_id = outer_frame_id;
> +      return;
> +    }
> +
>     /* Our frame ID for a stub frame is the current SP and LR.  */
>     arm_gdbarch_tdep *tdep
>       = gdbarch_tdep<arm_gdbarch_tdep> (get_frame_arch (this_frame));
> @@ -3725,6 +3743,11 @@ arm_m_exception_prev_register (frame_info_ptr this_frame,
>       *this_cache = arm_m_exception_cache (this_frame);
>     cache = (struct arm_prologue_cache *) *this_cache;
>   
> +  /* It's not allowed to call prev_register when this_id has returned the
> +     outer_frame_id.  The arm_m_exception_cache function will return NULL when
> +     the frame cannot be properly unwinded.  */
> +  gdb_assert (cache != NULL);
> +

It does seem safe to assume this function won't be called if there is no frame_id. So I agree this would be
a GDB bug and needs an assertion here.

>     /* The value was already reconstructed into PREV_SP.  */
>     arm_gdbarch_tdep *tdep
>       = gdbarch_tdep<arm_gdbarch_tdep> (get_frame_arch (this_frame));