Re: [Qemu-devel] [RFC PATCH] mmio-exec: Make device return MemoryRegion rather than host pointer

[Paolo Bonzini <pbonzini@redhat.com>]

On 26/04/2018 17:09, Peter Maydell wrote:
> 
> QUESTIONS:
>  * do we really need to postpone things from
>    memory_region_invalidate_mmio_ptr() to
>    memory_region_do_invalidate_mmio_ptr(), given that the deletion of
>    the memory region and ramblock is RCU-deferred anyway?
>  * should we add the subregion at elevated prio so it definitely hits
>    first?  I've left it the way the existing code does for the
>    moment...
>  * is there any way to avoid the weird self-owning MemoryRegion
>    and corresponding need to pass a NULL name pointer?

There would be, but the comment here says why it'd be an issue:

    /* Memory regions without an owner are supposed to never go away;
     * we do not ref/unref them because it slows down DMA sensibly.
     */
    if (mr && mr->owner) {
        object_ref(mr->owner);
    }

though perhaps we can remove that other hack in 2.13, which will have
MemoryRegionCache again.

Paolo