Re: [PATCH] dump_stack() based on prologue code analysis

[Franck Bui-Huu <vagabon.xyz@gmail.com>]

Atsushi Nemoto wrote:
> On Thu, 27 Jul 2006 21:03:07 +0200, "Franck Bui-Huu" <vagabon.xyz@gmail.com> 
> wrote:
>>>> +     info.func = (void *)(pc - ofs);
>>>> +     info.func_size = ofs;   /* analyze from start to ofs */
>> in get_frame_info(), there is the following condition to stop the
>> prologue analysis
>>
>>              if (info->func_size && i >= info->func_size / 4)
>>                      break;
>>
>> Setting info.func_size = ofs may trigger this stop condition very
>> early, specially if "ofs" is small...I would simply remove this
>> condition since it's very empirical and IMHO not very usefull.
> 
> Yes, that is what I wanted.  Imagine if a exception happened on first
> place on non-leaf function.  In this case, we must assume the function
> is leaf since RA is not saved to the stack.
> 

The only case I can imagine is when sp is corrupted which is unlikely.
However an exception can occure just after a prologue of a nested
function which is more likely. In that case you will assume wrongly
that the function was a leaf one.

I don't think we gain more than we loose with this test. Maybe we can
just leave

                if (i >= info->func_size)
                        break;

for safety purpose.

                Franck