在 2023/8/31 18:04, Jiaxun Yang 写道:
在 2023/8/31 17:42, Huacai Chen 写道:
On Thu, Aug 31, 2023 at 4:58 PM Jiaxun Yang jiaxun.yang@flygoat.com wrote:
在 2023/8/31 14:22, Huacai Chen 写道:
[...]
But it is worth to try, and you can walk the children to set mrrs when the quirk runs on bridges, I think.
No, this will break hotplug, I had managed to get hotplug work on 2K1000. Also we have no guarantee on order of discovering devices.
but the old quirk should run on every single device.
Your current patch has a drawback that both quirks will run for MIPS, and their order is random (though it may cause nothing, but not elegant).
Actually loongson_mrrs_quirk is declared by DECLARE_PCI_FIXUP_EARLY but loongson_old_mrrs_quirk is declared by DECLARE_PCI_FIXUP_ENABLE, which means loongson_old_mrrs_quirk always runs after loongson_mrrs_quirk as expected.
Then I think it is better to put this quirk in arch/mips/loongson64/, just as pci_fixup_radeon().
Huacai, if you are happy with changes in this patch, do you mind to give out a conditional R-b tag?
Since this regression is in tree for a while, I really don't want to see it lapse into next release.
Hmm, spiting two quirks with similar functionality into two places doesn't sound like a good idea.
PCI folks, what do you think?
Bjorn, Lorenzo and Krzysztof, do you mind to give some input on this fix?
Thanks - Jiaxun
Thanks.
- Jiaxun
[...]