On Mon, Aug 15, 2016 at 06:35:03PM -0400, Keith Busch wrote: > On Mon, Aug 15, 2016 at 02:50:58PM -0500, Bjorn Helgaas wrote: > > What? This is August 15, not April 1. That makes no sense > > whatsoever. > > Yes, I agree, and not advocating this was a good idea. I'm just the > messenger in this case. > > That said, it looks like this can fall under the intended usage for > the quirk framework if we can consider this a work-around for standard > non-compliance. Or are your thoughts that we've gone too far if it's > not a true erratum? Well, I don't want to add random quirks for everybody who decides to make their own non-standard hardware. The reason we have the PCI specs is for interoperability, and this is not it. The whole purpose of capability bits is for the hardware to tell the OS that it has certain *capabilities* that work in certain ways. It's idiotic to set a capability bit that says "I have this indicator" when it doesn't work the way it's supposed to. The hardware should just NOT SET THE BIT. How hard is that? I can't conceive of the reasoning that concluded this was useful. I don't care if you want to make the LEDs do something else, but for Pete's sake, don't claim they work according to spec when they don't. I assume there will be more broken devices like this in the future. How long will we be adding quirks? Do you have a plan to resolve this the correct way eventually? I still don't know what it means that the devices "do the wrong thing." Do the LEDs not work? Do the LEDs work according to the PCIe spec, but the patterns don't match the International Blinken Lights spec? Does the hotplug controller lock up? Sigh. -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
