Re: [PATCH] agp/amd64: Bind to unsupported devices only if AGP is present

[Hans de Goede <hdegoede@redhat.com>]

Hi,

On 25-Jun-25 4:33 PM, Lukas Wunner wrote:
> On Wed, Jun 25, 2025 at 04:08:38PM +0200, Hans de Goede wrote:
>> Lukas made me aware of this attempt to fix the KERN_CRIT msg, because
>> I wrote a slightly different patch to fix this:
>>
>> https://lore.kernel.org/dri-devel/20250625112411.4123-1-hansg@kernel.org/
>>
>> This seems like a cleaner fix to me and something which would be good
>> to have regardless since currently the driver_attach() call is doing
>> too much work because the promisc table catches an unnecessary wide
>> net / match matching many PCI devices which cannot be AGP capable
>> at all.
> 
> So how do you know that all of these unsupported devices have
> PCI_CLASS_BRIDGE_HOST?

The top of the driver says

 * This is a GART driver for the AMD Opteron/Athlon64 on-CPU northbridge.
 * It also includes support for the AMD 8151 AGP bridge

Note this only talks about north bridges.

Also given the age of AGP, I would expect the agp_amd64_pci_table[]
to be pretty much complete and the need for probing for unknown AGP
capable bridges is likely a relic which can be disabled by default.

Actually the amd64-agp code is weird in that has support for
unknown AGP bridges enabled by default in the first place.

The global probe unknown AGP bridges bool which is called
agp_try_unsupported_boot is false by default.

As discussed in the thread with my patch, we should probably
just change the AMD specific agp_try_unsupported to default
to false too.

> The only thing we know is that an AGP
> Capability must be present.
> 
> In particular, AGP 3.0 sec 2.5 explicitly allows PCI-to-PCI bridges
> in addition to Host-to-PCI bridges.

Ok, so we can add a second entry to the agp_amd64_pci_promisc_table[]
to match PCI to PCI bridges just to be sure, that still feels
cleaner to me.

Regards,

Hans