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

[David Daney <ddaney@avtrex.com>]

Nigel Stephens wrote:
> 
> 
> David Daney wrote:
> 
>> Atsushi Nemoto wrote:
>>
>>> On Thu, 27 Jul 2006 20:12:45 +0100, Thiemo Seufer <ths@networkno.de> 
>>> wrote:
>>>
>>>> IOW, binary analysis can't be expected to provide full accuracy, but
>>>> we can live with a reasonable approximation, I think.
>>>
>>>
>>>
>>> Yes, this is a starting point.
>>>
>>> The patch (and current mips get_wchan() implementation) tries to do is
>>> what I used to do to analyze stack dump by hand.
>>>
>>> 1. Determine PC and SP.
>>> 2. Disassemble a function containing the PC address.
>>> 3. If the function is leaf, make use RA for new PC.
>>
>>
>> This was always the tricky part for me.  How do you know if the 
>> function is a leaf?
>>
> 
> I think that if you cannot find a store instruction which saves RA to 
> the stack -- either because it's a real leaf and there is no such store, 
> or because the PC hasn't yet reached the store instruction -- then in 
> both cases it can be treated as a leaf.

Presumably you are walking the code back from the PC until you find the 
prolog.  How would you tell if you had gone past the beginning of a leaf 
function?  If you find a j $31 you might assume that it was the end of 
the previous function.

But that does not work if you are in a function that has multiple return 
points.  On encountering a j $31 you have no way of telling if you are 
in a leaf function or a non-leaf function with multiple return points.

I may be missing something here, if you know of a failure-proof manner 
to detect leaf functions I would appreciate hearing what it is.

Thanks,
David Daney.