Re: [RFA/RFC] ARM : one-line prologue analysis

[Richard Earnshaw <rearnsha at arm dot com>]

> >     for (reg = 0; reg < NUM_REGS; reg++)
> > !     if (cache->saved_regs[reg].addr != -1)
> >         cache->saved_regs[reg].addr += cache->prev_sp;
> 
> Would trad_frame_addr_p() be better?
> 
> Andrew
> 
> 

I think so.  However, arm_scan_prologue (which is run just before this 
loop) abuses the .addr field and puts an offset into it.  That offset can 
be (often is) negative at that point; the loop Jerome is fixing is the one 
where that offset is turned back into a real address.

Fortunately, the stack will always be word aligned (your application is 
really broken if it isn't), and registers are always at least four bytes 
in size, so I don't think it's possible for -1 to ever get written in 
there by the code itself.

So approved with Andrew's suggested change.

R.

PS Andrew,

We seem to be horribly overloading and then exposing these special values 
in the trad_frame interface.  Wouldn't it make more sense to add more 
routines to get and update the values, then we can hide the special 
meanings entirely (and maybe even change the way they are recorded if we 
need to).

R.