[m68k]: patch to fix m68k_output_mi_thunk() for ColdFire

[Peter Jakubek]

Mark Mitchell wrote:
>> Bernardo Innocenti wrote:
>>
>> Mark, I believe this patch is also needed on 3.4.
>> Without it, GCC may output invalid code for
>> ColdFire targets.
>>
>> I can't find a specific PR, but I may file one if needed.
>
> Please file a PR demonstrating the problem and attach the patch
> to it (or include a URL), and target the patch for 3.4.1.
>

Yes, the patch is needed for gcc 3.4 as well. here is code that
allows to reproduce the problem.
target m68k-elf, compile with -m5307:

     class FooA {
     public:
       virtual ~FooA () = 0;
       int data[100];
     };


     class FooB {
     public:
       virtual void foo1 () = 0;
               void fooB () {foo1 ();}
       int data[100];
     };


     class FooC : public FooA, public FooB {
     public:
       virtual void foo1 () {}
     };


     FooC* Test ()
     {
       FooC* foc = new FooC;
       foc->fooB ();
     }

The thunk comes into play when FooB::fooB () calls the virtual
function foo1. While in FooB::fooB () the this pointer points
to FooB. Then when foo1 is called (which is overwridden by
FooC::foo1), the this pointer needs to be adjusted to point
to FooC, which is -404 in this example (the size of FooA).

Without the patch gcc generates "add.l #-404,4(%sp)" which is
not supported by ColdFire and rejected by the assembler.

I hope this helps,

Peter Jakubek
LaserAnimation Sollinger GmbH
peter@lasergraph-dsp.de