Re: [Qemu-devel] [PATCH 4/5] Report exact PC on watchpoint hit

[Paul Brook]

On Saturday 31 May 2008, Jan Kiszka wrote:
> Paul Brook wrote:
> >> @@ -2395,6 +2396,11 @@ static target_ulong check_watchpoint(tar
> >> +                if (env->current_tb) {
> >> +                    tb_addr = env->current_tb->page_addr[0];
> >> +                    tb_invalidate_phys_page_range(tb_addr,
> >> +                                            tb_addr+TARGET_PAGE_SIZE-1,
> >
> > This is wrong. env->current_tb is not what you think it it is. TB
> > chaining means we may be an any TB reachable from there.
>
> OK, I see. But the general approach to obtain the exact PC is fine? So
> should I flush the whole memory range instead?

By whole memory range you mean all of everything? That sounds a fairly poor 
solution.  The way arm and m68k handle this is to never put a memory op in 
the middle of a TB.

Your solution also only works for targets that define TARGET_HAS_PRECISE_SMC.

Paul