Re: [Qemu-devel] [PATCH 08/12] pci: allow 0 address for PCI IO regions

[Michael Roth]

Quoting Claudio Fontana (2015-01-09 08:57:39)
> Hello,
> 
> resurrecting an old thread.. I incurred in the same issue being
> discussed before,
> where QEMU silently ignores PCI BAR address programming attempts where
> the I/O space offset is 0 (zero).
> 
> I think that from a QEMU "user" standpoint, beside this particular issue,
> which can be easily worked around just using a minimum offset,
> it would be good if QEMU would be a bit verbose in producing a warning
> about this.
> 
> I think that at least it would be worth a message if DEBUG_PCI is set.
> 
> This silent discarding of BAR programming attempts has been painful
> while doing early enablement
> even for other cases (like the requirement to set I/O space bit before
> hand etc), which are legitimate,
> but are still worthy of a diagnostic I think, at the very least if
> doing pci enablement (which to me translates in having DEBUG_PCI set).
> 
> What do you guys think, would a patch be welcome trying to address that?
> Would you make the diagnostic dependent on DEBUG_PCI?

I've sent an updated patch (which allows 0-address MEM regions as well)
as a separate patchset. My hope is that we can simply allow the
programming of 0-address mem/io bars, but there are some concerns I
still don't quite understand but have attempted to summarize to
continue the discussion:

http://lists.gnu.org/archive/html/qemu-devel/2014-12/msg03453.html

Debugging would be useful, but there are undoubtedly cases where the
current behavior prevents proper initialization of guest devices on
existing guests which expect 0 to be valid, so at the very least I
think we should allow the behavior to be enabled on a
machine/host-bridge level of granularity, if not for all
machines/host-bridges as the patch above does.

> 
> Thanks,
> 
> Claudio