On Thu, 2024-12-12 at 10:08 +0100, Lukas Wunner wrote: > On Wed, Dec 11, 2024 at 03:07:38PM +0200, Ilpo Järvinen wrote: > > I recall taking note of this inconsistency in some lspci dumps I've from > > Mika (but forgot it until now). So I'm afraid it might be more widespread > > than just TR. > > Oh you did? Interesting. After re-reading the spec I'm convinced now > that we're doing this wrong and that we should honor the Max Link Speed > instead of blindly deeming all set bits in the Link Capabilities 2 > Register as supported speeds: > > https://lore.kernel.org/r/e3386d62a766be6d0ef7138a001dabfe563cdff8.1733991971.git.lukas@xxxxxxxxx/ > > @Niklas, could you test if this is sufficient to avoid the issue? > Or do we still need to stop instantiating the bandwidth controller > if more than one speed is supported? > > Thanks! > > Lukas Yes, I will test this but will only get to do so tonight (UTC +2). If it's not sufficient I think we could use the modified pcie_get_supported_speeds() to check if only one link speed is supported, right? Thanks, Niklas
