Re: [Qemu-devel] [PATCH 2/2] m48t59: add mem_base value to m48t59_init_isa()

[Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>]

On 19/01/15 15:01, Andreas Färber wrote:

> Am 19.01.2015 um 13:57 schrieb Artyom Tarasenko:
>> On Mon, Jan 19, 2015 at 1:45 PM, Paolo Bonzini <pbonzini@redhat.com> wrote:
>>> On 19/01/2015 12:35, Mark Cave-Ayland wrote:
>>>> Similar to m48t59_init(), add a mem_base value so that NVRAM can be mapped via
>>>> MMIO rather than ioport if required.
>>>>
>>>> Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
>>>> ---
>>>
>>> Is it really ISA if it's MMIO?  In other words, why can't this be a
>>> sysbus device?
>>
>> On physical machines it's EBus, which is pretty much like 8-bit ISA.
>> So, I think modelling it as ISA is closer to to the reality.
>> But out of curiosity, would it be possible to have a sysbus device
>> somewhere in a middle of PCI space? [...]
> 
> Why would you want to use a SysBusDevice in the first place? I
> previously discussed with Mark that it should be an EBusDevice, not an
> ISADevice or SysBusDevice. IndustryPack is an example of a custom bus
> that sits behind a PCI bridge and doesn't need a global variable.

I can see this makes logical sense - I guess the reason it hasn't been
done was to avoid having to write EBus-specific initialisation code for
every device which would only be used on one platform. So you're
suggesting that IndustryPack is a way of doing this?

> Also, wasn't Hervé's(?) plan to get rid of mem_base completely by always
> passing a pointer to ISADevice/ISABus around? It should only be needed
> when somewhere NULL is being passed, no?

That would definitely be better for wiring things up with -device but
wouldn't that involve changing all of the existing ISA devices?


ATB,

Mark.