Re: [Qemu-devel] BUG: QEMU aborts when setting breakpoint in gdb (bisected)

[Michael S. Tsirkin]

On Wed, Nov 06, 2013 at 10:13:16PM +0100, Paolo Bonzini wrote:
> Il 06/11/2013 19:39, Michael S. Tsirkin ha scritto:
> > Because this will affect performance in unpredicatable way.
> 
> It's not really unpredictable.  It can be easily unit-tested,

Go ahead, post, test, I'm not stopping you.

> and anyway
> the targets with 64-bit address spaces don't have any particular
> performance problem.

That's only s390x and they happen not to use IO for anything,
instead they do hypercalls. So hard to tell.

> > We can't make such changes in 1.7 IMHO:
> > it would need much more than just a quick "works for me".
> 
> I can say the same about phys_page_find.

Why? It's *obvious* this code is not designed to work with
addresses > target address space.
You doubt that?
What is the worry here?

>  It's been obviously broken for
> years and nobody ever cared, it cannot be super-urgent now to fix it.
> 
> Paolo

Well we added a bunch of code and now it is failing.

> >> >  I don't feel confident
> >> > changing phys_page_find, even if it's just 2 lines.
> >> > 
> >> > Paolo
> > Well it's *obviously* broken if address is outside target address
> > space.
> > Take a look at the patch first, then argue.