Re: [Qemu-devel] [PATCH target-arm v1 1/1] arm/xilinx_zynq: Always instantiate the GEMs

[Peter Maydell]

On 3 January 2014 16:56, Peter Crosthwaite <address@hidden> wrote:
> On Sat, Jan 4, 2014 at 2:22 AM, Peter Maydell <address@hidden> wrote:
>> On 3 January 2014 16:15, Peter Crosthwaite <address@hidden> wrote:
>>> One subtle question before respinning - should the
>>> qdev_set_nic_properties actually be conditional? Setting it to an
>>> unused NIC should be harmless, so perhaps it should be outside the
>>> if(nd->used). Other boards/MAC combos that dont do the
>>> qdev_check_nic_model can then just dispose of the if (nd->used) check
>>> altogether.
>>
>> Well qdev_set_nic_properties as it stands today will unconditionally
>> assume you've passed it a valid NICInfo, so we have to a void calling
>> it if nd->used isn't set.
>
> So I guess I'm thinking that the unusued nics should be a "valid"
> NICInfo - just one that doesn't connect to anything. Currently,
> qdev_set_nic_properties() sets three props - "mac", "netdev" and
> "vectors". "mac" looks dubious to me, I'm not sure why the net layer
> is telling devices what their MAC addresses are - it should be the
> other way round.

The device doesn't (in the old -net model) know what its MAC address
is, because that's something you specify in the -net command line option.
So this is the passing on of that info to the device.

> Setting of "netdev" is already protected against NULL
> setting so that one is handled. "vectors" looks like it has a sanity
> check guard on its setting as well via the DEV_NVECTORS_UNSPECIFIED
> check. Although that value may rely on some net.c code paths for
> non-NULL NICs.

I think it would probably be better to look into properly cleaning up
the legacy -net nic support and moving embedded devices over to
work with some variant of the new command line model, rather
than tweaking the legacy stuff. nd_table eventually should ideally
just go away I think.

thanks
-- PMM