Re: [Qemu-devel] [PATCH 0/3] block: Avoid direct AIO callback

[Kevin Wolf <kwolf@redhat.com>]

Am 10.06.2011 17:32, schrieb Luiz Capitulino:
> On Tue,  7 Jun 2011 16:18:30 +0200
> Kevin Wolf <kwolf@redhat.com> wrote:
> 
>> This series fixes some cases of block drivers calling AIO callbacks too early.
>> It fixes the IDE assertion failure reported by Luiz (in error cases, the DMA
>> status, including acb, could first be reset in the callback and only then be
>> set by the caller, resulting in a dangling acb and wrong status register value).
> 
> This fixes the reported bug, thanks.
> 
> I know this is a different subject, but I'm still unable to use the host cdrom
> if the -snapshot flag is passed, I think the idea of ignoring the flag for a
> read-only device would fix this, no?

Yes, it would. But as we discussed it would have other implications that
I wouldn't feel comfortable about (surprising semantics of 'commit'
would be one).

Passing things like bdrv_eject() or bdrv_is_inserted() to the backing
file still sounds like a cleaner approach, but probably isn't as easy.

Kevin