Re: [Qemu-devel] Re: [patch] gdb-stub support for Alpha

[Vince Weaver <vince@csl.cornell.edu>]

On Thu, 4 Dec 2008, Jan Kiszka wrote:
> I would suggest checking out what gets translated, executed and
> communicated by using '-d in_asm,out_asm' for the qemu command line,
> running qemu itself inside gdb and maybe also enabling 'set debug remote
> 1' in the remote gdb instance (which will print the frontend<->backend
> communication). That should help to get a better picture about what
> happens and why you see double steps.

Thanks!  That helped track down the problem.

On Alpha, for non-branch instructions the move of ctx->pc to cpu_pc 
happens at the end of the TB (after the single-step exception is called).

However, for branches, the move of ctx->pc to cpu_pc happens within the 
instruction decoding, *before* the single-step exception is called, so the 
exception handler over-writes the proper new pc with an older one.

This is tricky to fix.  Either all the branch code needs to be re-written 
to write to a temporary nextpc value that is written at the end, or else
I have to somehow wrap the single-step exception code to preserve the new 
pc value.

Vince