On Wed, 2011-06-29 at 00:36 -0500, Jon Mason wrote: > > > > - I would like a way to specify the MPS of the host bridge, it may > > > > not be the max of the RC P2P bridge (it -generally is- but I'd like a > > > > way to override that easily). > > > > > > I could do another boot arg for this. > > > > Please no... not boot args. Those suck. Besides things can be different > > per host bridge, especially on !x86 platforms. Best is pcibios_* helpers > > (that take the bus as an argument) which the arch can populate for these > > things. > > Ok, no boot arg. Well, I suspect the x86 crowd will probably still want one :-) IE. The bootarg is going to come handy to diagnose problems if/when this patch causes some. It just think we should come up with a reasonable default (which should be either the "relaxed" way imho or arch set because I want powerpc to default to that) and a boot arg to override the arch. Forget about the helper I just wrote about, it can be global, I think your idea of me using the RC bridge to "fixup" the values there is good (as long as there's a quirk I can use to do so between you reading it from the device and later using it to fixup the devices). > It's possible to have any arch code specify the > pcie_mpss prior to the buses being walk, which would work around this > issue. Correct, we could do that. > This might be a bit hacky. If it ends up being too difficult, > we can do something similar to what you suggest above. No I think having the arch manipulate the RC from a quirk if it wants to is fine. The main question is which of the "conservative" vs. "relaxed" mode shall be the default. > > > > - I think we need to handle MRRS at the same time. We need MRRS to be > > > > clamped by MPS, don't we ? In addition to being clamped top-down. > > > > > > You are 100% correct. I was working on a follow-on patch to do it, > > > but I'll just put it into here. > > > > Ok. > > This is implemented in the patch I just sent out. Please look it > over. Well, our IRC discussion seem to have gone the other way :-) It looks like we are better off leaving MRRS alone with this patch and tackling it with a separate one. I tend personally to prefer making sure MRRS <= MPS as a sane "default" at boot time, and let the driver do whatever it wants (or even expose it via sysfs). But we shall make it a separate patch at least for the sake of bisectability in case something in our reasoning is wrong. Cheers, Ben. -- 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
