Re: [Qemu-trivial] [Qemu-devel] [PATCH] serial-pci: remove memory regions from BAR before destroying them

[Paolo Bonzini <pbonzini@redhat.com>]

Il 14/07/2014 14:36, Peter Crosthwaite ha scritto:
> On Thu, Jun 26, 2014 at 5:32 PM, Markus Armbruster <armbru@redhat.com> wrote:
>> Paolo Bonzini <pbonzini@redhat.com> writes:
>>
>>> Otherwise, hot-unplug of pci-serial-2x trips the assertion
>>> in memory_region_destroy:
>>>
>>>     (qemu) device_del gg
>>>     (qemu) qemu-system-x86_64: /work/armbru/tmp/qemu/memory.c:1021: memory_region_destroy: Assertion `((&mr->subregions)->tqh_first == ((void *)0))' failed.
>>>     Aborted (core dumped)
>>>
>>> Reported-by: Markus Armbruster <armbru@redhat.com>
>>> Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
>>
>> Reviewed-by: Markus Armbruster <armbru@redhat.com>
>>
>
> Reviewed-by: Peter Crosthwaite <peter.crosthwaite@xilinx.com>
>
>> Would it make sense to add a "must not contain subregions" to
>> memory_region_destroy()'s function comment?
>>
>
> Any reason to just not patch the memory region finaliser to unparent

Note that unparent for memory regions is _not_ 
memory_region_del_subregion.  It is memory_region_destroy.

The parent object of a memory region is a device; the _container_ of a 
memory region is another memory region.

> all contained subregions automatically rather than assert? Destroying
> a container should imply removing the subregion relationship and
> simply orphan the subregion.

This makes sense since we will soon make memory_region_destroy optional 
(devices will automatically destroy their memory regions).  Before 
QOMification, however, I think the assert was a useful debugging tool, 
guaranteeing that owners of memory regions were destroyed in the right 
order.  So we could indeed revisit this in 2.2 and make 
memory_region_del_subregion also optional.

Paolo