On Tue, Sep 6, 2011 at 11:52 AM, Benjamin Herrenschmidt <benh@xxxxxxxxxxxxxxxxxxx> wrote: > On Tue, 2011-09-06 at 11:11 -0700, Jon Mason wrote: >> Yes, that is why I added the MRRS setting code originally. Since the >> MRRS setting is almost universally larger than the MPS and some >> devices are freaking when their MRRS is changed, we must either use >> the "safe" option by default and not touch the MRRS or allow the >> "performance" option to have this hole. I suppose a third option is >> to create a list of faulty devices that cannot handle MRRS >> modification, but that seems rather ugly. > > Do we know for sure that card can't handle it at all or only freaks out > with "0" ? It's hard to say since the MPS on most systems is "0" (e.g., 128B). It appears that one of the Radeon chips doesn't like the MRRS set to 0/6/7 (according to a patch attached to the same bug). So, there are a number of invalid settings that will have a negative impact on that. If we don't want to allow the hole in the patch I pushed to Jesse, I think we should go with the "safe" option not setting the MRRS and make it the default. > IE Does the card work if MRS == MPS of the card (ie using my algorithm > and assuming the host bridge supports arbitrary MPS). Currently, the code will set the MRRS to the MPS of the parent (if the "performance" option is used). It will then read back the value to verify that it has been changed. If not, then it will try setting it to the next value lower, etc. until it hits 0. The "safe" option tries to set the MRRS to a uniform size to make everything equitable, which seemed logical but not 100% necessary. I am more than happy to have this option not touch the default MRRS at all. >> I'm open to any option, but there seems to be a large number of issues >> with the MRRS modification code and we need to get a fix into 3.1. > > Ok, I only head of one, I didn't know there were more. I've gotten 5 e-mails so far. I'm betting there are more out there. > Are these typically hitting with the "performance" option ? IE. It make > sense to leave MRRS untouched in the "safe" case. The patch I sent out still used the "performance" option without modifying the default MRRS of the device. All that have tested it said that it resolves their issues. thanks, Jon > > 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
