On 3/25/19 12:13 PM, Heiner Kallweit wrote: > On 25.03.2019 19:44, Florian Fainelli wrote: >> On 3/25/19 11:35 AM, Heiner Kallweit wrote: >>> 1000BaseT standard requires that a link is reported as down earliest >>> after 750ms. Several use case however require a much faster detecion >>> of a broken link. Fast Link Down supports this by intentionally >>> violating a the standard. This patch exposes the Fast Link Down >>> feature of 88E1540 and 88E6390. These PHY's can be found as internal >>> PHY's in several switches: 88E6352, 88E6240, 88E6176, 88E6172, >>> and 88E6390(X). Fast Link Down and EEE are mutually exclusive. >>> >>> Signed-off-by: Heiner Kallweit <hkallwe...@gmail.com> >> >> This looks fine, just one question though: do not you need to verify >> that the phy_interface_t maps to a copper medium somehow? >> > Good question. The register we configure is used in copper mode > only (see name: COPPER_CTRL3). So we don't cause any harm at least > if we configure Fast Link Down when being, let's say, in a fiber mode. > Fast Link Down is applicable for 1000BaseT only, so in theory what > you say would apply also to an attempt to configure Fast Link Down > when being in 100BaseT or 10GBaseT copper modes. > Being allowed to configure a feature for 1000BaseT whilst being in > a different mode can been seen as a feature (what I do) or as a bug. > Most likely there is no 100% answer to this question. >
Agree, and the reasoning about the register we are reading/writing to is indeed not a problem, it is more about allowing the tunable to be configured or not, and in fact, this is a policy check that can be done above the PHY driver itself. -- Florian
