[PATCH v2 1/1] Split VGA default device handler out of VGA arbiter

[Ard Biesheuvel]

On 22 August 2017 at 23:19, Bjorn Helgaas <helgaas at kernel.org> wrote:
> On Mon, Aug 21, 2017 at 11:53:01AM +0100, Lorenzo Pieralisi wrote:
>> On Thu, Aug 17, 2017 at 09:30:28PM +1000, Daniel Axtens wrote:
>> > A system without PCI legacy resources (e.g. ARM64) may find that no
>> > default/boot VGA device has been marked, because the VGA arbiter
>> > checks for legacy resource decoding before marking a card as default.
>>
>> I do not understand this paragraph, in particular:
>>
>> - "A system without PCI legacy resources (e.g. ARM64)". What does this
>>   mean ? I take this as "ARM64 does not support IO space"; if a PCI host
>>   bridge supports IO space, there is nothing from an architectural
>>   point of view that prevents an MMIO based IO space implementation to
>>   work on ARM64. It is PCI bridge specific, not arch specific.
>
> This reference to ARM64 is the same thing I stumbled over.  Maybe it
> could be written along the lines of:
>
>   The VGA arbiter selects a default VGA device that is enabled and
>   reachable via the legacy VGA resources (mem 0xa0000-0xbffff, io
>   0x3b0-0x3bb, io 0x3c0-0x3df, etc).
>
>   If there is no such device, e.g., because there's no enabled VGA
>   device, the host bridge doesn't support access to those legacy
>   resources, or a PCI-PCI bridge doesn't have VGA Enable set, a
>   platform may select an arbitrary device by calling
>   vga_set_default_device().
>
> Then I think this patch changes the previous behavior by allowing the
> arbiter to select a device that is enabled and has a driver but may
> not be reachable via the legacy VGA resources.
>
> I don't know what all the consequences of that would be, but it does
> muddy the VGA arbiter waters a bit.  AIUI, the main reason for the
> arbiter is to allow multiple legacy VGA devices by manipulating bridge
> VGA Enable bits so only one receives the legacy resources at a time.
>
> These devices that do not need the legacy resources don't need that
> special treatment because they don't care about the bridge VGA Enable
> bits and several can coexist in the system with no need for VGA
> arbitration.
>
> I guess the powerpc fixup_vga() already selects a device that doesn't
> need the VGA resources, so we already have that situation.
>

Perhaps it is unclear that the whole point of tinkering with the VGA
arbiter is that X may refuse to use a GFX card if it is not tagged as
the default VGA device by the arbiter. I have suggested before that
fixing X may be more appropriate in this case, given that ownership of
the legacy VGA resources may not correlate at all with being the
primary display device on non-x86 architectures.