On Tue, May 30, 2023 at 05:27:24PM -0500, Bjorn Helgaas wrote: > Ah, you're right, sorry I missed that. Dispensing with the SERDES > details would make this more obvious. Lesson learned. When I had just gotten out of college, every time I asked the coworkers in my company what they're up to, I was amazed by them just proceeding to tell me all the nitty gritty details of what they're doing and debugging, like I was supposed to understand or care for that matter. "Dude, can't you just paint the high level idea without using dorky words?" Now I'm one of them... > Not sure why this needs to change the pci_scan_slot() path, since > Function 0 is present and enumerable even though it's not useful in > some cases. Well, the rationale for me was pretty simple: it's the pci_scan_slot() logic that I want to change - continue enumeration in some cases when the pci_dev for fn 0 is NULL - and I'm otherwise perfectly okay with pci_scan_slot() getting a NULL pci_dev from pci_setup_device() for fn 0. That wasn't something I had in mind to change. This patch is what it takes to propagate a qualifier, without leaving a mark in any structure, for that NULL return code: is it NULL because enumeration came up with nothing, or is it NULL because pci_set_of_node() said so? > Seems like something in pci_set_of_node() or a quirk could do whatever > you need to do. Could you help me out with a more detailed hint here? I'm not really familiar with the PCI core code. You probably mean to suggest leaving a stateful flag somewhere, though I'm not exactly sure where that is, that would reach pci_scan_slot() enough to be able to alter its decision.
