On Wed, 25 Mar 2020 10:11:12 +0000 Russell King - ARM Linux admin <linux@xxxxxxxxxxxxxxx> wrote: > On Wed, Mar 25, 2020 at 09:34:49AM +0100, Oleksij Rempel wrote: > > Hi all, > > > > I'm working on mainlining of NXP1102 PHY (BroadR Reach/802.3bw) support. > > > > Basic functionality is working and support with mainline kernel. Now it is > > time to extend it. According to the specification, each PHY can be master > > or slave. > > > > The HW can be pre configured via bootstrap pins or fuses to have a default > > configuration. But in some cases we still need to be able to configure the > > PHY in a different mode: > > -------------------------------------------------------------------------------- > > http://www.ieee802.org/3/1TPCESG/public/BroadR_Reach_Automotive_Spec_V3.0.pdf > > > > 6.1 MASTER-SLAVE configuration resolution > > > > All BroadR-Reach PHYs will default to configure as SLAVE upon power up or > > reset until a management system (for example, processor/microcontroller) > > configures it to be MASTER. MASTER-SLAVE assignment for each link > > configuration is necessary for establishing the timing control of each PHY. > > > > 6.2 PHY-Initialization > > > > Both PHYs sharing a link segment are capable of being MASTER or SLAVE. In > > IEEE 802.3-2012, MASTER-SLAVE resolution is attained during the > > Auto-Negotiation process (see IEEE 802.3-2012 Clause 28). However, the > > latency for this process is not acceptable for automotive application. A > > forced assignment scheme is employed depending on the physical deployment > > of the PHY within the car. This process is conducted at the power-up or > > reset condition. The station management system manually configures the > > BroadR-Reach PHY to be MASTER (before the link acquisition process starts) > > while the link partner defaults to SLAVE (un-managed). > > -------------------------------------------------------------------------------- > > > > Should phylink be involved in this configuration? What's the proper user > > space interface to use for this kind of configuration? ethtool or ip > > comes into mind. Further having a Device Tree property to configure a > > default mode to overwrite the boot strap pins would be nice to have. > > Well, the first question would be whether this is something that > userspace needs to alter, or whether static configuration at boot > time is what is necessary. > > Given what is in the description, it seems that the concern is with > the latency it takes for the link to come up. I would suggest that > the lowest latency would be achieved when using static configuration > rather than waiting for the kernel to fully boot and userspace to > start before configuring the PHY. Yes, that would be the fastest, and in many cases the preferred way. But the lack of auto negotiation is not a choice. It is imposed by the spec. Because of this, and since the PHY's are configurable in software, there is some need for configuration in user-space. Of course latency would not be an issue in such a case, otherwise a fixed strapped configuration was chosen. Best regards, -- David Jander Protonic Holland.
