[OpenBIOS] [patch 6/6] Prevent dereferencing page zero and complain about invalid handles

[Blue Swirl]

On Fri, Aug 21, 2009 at 11:23 PM, Igor
Kovalenko<igor.v.kovalenko at gmail.com> wrote:
> On Fri, Aug 21, 2009 at 11:12 PM, Blue Swirl<blauwirbel at gmail.com> wrote:
>> On Fri, Aug 21, 2009 at 12:16 AM, Igor
>> Kovalenko<igor.v.kovalenko at gmail.com> wrote:
>>> Index: openbios-devel/forth/device/package.fs
>>> ===================================================================
>>> --- openbios-devel.orig/forth/device/package.fs
>>> +++ openbios-devel/forth/device/package.fs
>>> @@ -59,10 +59,15 @@
>>>
>>>  : find-method ( method-str method-len phandle -- false | xt true )
>>>   \ should we search the private wordlist too? I don't think so...
>>> -  >dn.methods @ find-wordlist if
>>> -    true
>>> +  ?dup if
>>> +    >dn.methods @ find-wordlist if
>>> +      true
>>> +    else
>>> +      2drop false
>>> +    then
>>>   else
>>> -    2drop false
>>> +      cr ." find-method: " type ."  : NULL phandle" cr
>>> +      false
>>>   then
>>>  ;
>>>
>>> @@ -75,10 +80,18 @@
>>>
>>>
>>>  : $call-method  ( ... method-str method-len ihandle -- ??? )
>>> -  dup >r >in.device-node @ find-method if
>>> -    r> call-package
>>> +  \ check if my-self exists; if not, there is nothing to call from
>>> +
>>> +  ?dup if
>>> +    dup >r >in.device-node @ find-method if
>>> +      r> call-package
>>> +    else
>>> +      cr ." $call-method: instance method not found" cr
>>> +      -821 throw
>>> +    then
>>>   else
>>> -    -21 throw
>>> +    cr ." $call-method: " type ."  : NULL ihandle" cr
>>> +    -f21 throw
>>>   then
>>>  ;
>>
>> Can we find out the caller in these cases? The error messages are not
>> very useful to the user. They also happen on Sparc32.
>>
>
> Eventually all callers would be covered, it takes time to trace
> forth code calls to origins. I think it is better to have extra messages
> in place while doing so.

Maybe the messages should be conditional to DEBUG_METHOD or something?

> In fact for me it takes so much extra time that I'm inclined to reimplement
> some forth code as C equivalents to keep debugging manageable.

Right, I wish there were equivalents of gdb breakpoints and commands
'where', 'up', 'down'. Doesn't rstack contain the return addresses?
Some code pushes other stuff to rstack, though.