| Message ID | 20171215002850.27862-1-f.fainelli@gmail.com |
|---|---|
| Headers | show |
| Series | net: dsa: Plug in PHYLINK support | expand |
On Thu, Dec 14, 2017 at 04:28:48PM -0800, Florian Fainelli wrote: > Hi all, > > This patch series replaces the existing PHYLIB integration with PHYLINK which > is a superior solution since we need to support a collection of fixed links, > integrated PHYs, SFP, non-pluggable SFPs etc. > > I am expecting quite a lot of breakage, for a number of reasons: > > - PHYLINK does not create a PHY device for fixed links (MLO_AN_FIXED), which > means that user-facing port (not DSA, not CPU) need to be explicitly handled > with phylink_mac_ops now > > - only been able to test this on a limited number of platforms, and not the > ZII devel B/C boards yet > > Please test and report! The good news is that it almost works on the ZII Rev C with the SFF modules, but there's a few issues that need solving: 1. s/WARN_ON(!lockdep_rtnl_is_held())/ASSERT_RTNL()/ in phylink.c - probably wouldn't have been noticed if it wasn't for the lockdep splat with DSA's interrupt handling during boot that's been there since -rc1. (Andrew mentioned that it's known, so I haven't reported that.) 2. ports without a phy but with sfp cause the "no phy at" and "failed to connect to port" message - I think we need a way for DSA to know whether it should be attempting to attach a PHY. 3. We need mv88e6xxx to implement at least phylink_mac_link_state() so that SFPs report sane ethtool information. It also needs to implement phylink_mac_config() so that the link can be appropriately configured to the module's capabilities, and phylink_validate() to do the jiggery for choosing between 1000base-X vs 2500base-X mode if the DSA switches can't automatically select the appropriate one. With a few hacks for the above, I can get TX_DISABLE on the SFF2 module to follow the interface up/down status, but as I say more work is needed so that mv88e6xxx works with the phylink callbacks for the SFF modules to be properly supported.
> 1. s/WARN_ON(!lockdep_rtnl_is_held())/ASSERT_RTNL()/ in phylink.c - > probably wouldn't have been noticed if it wasn't for the lockdep > splat with DSA's interrupt handling during boot that's been there > since -rc1. (Andrew mentioned that it's known, so I haven't > reported that.) Hi Russell https://lkml.org/lkml/2017/12/2/260 This should fix that splat. I hope it will get picked up at some point. It is probably getting on time to repost it. Andrew
Hi Den 15. des. 2017 01:28, skrev Florian Fainelli: > Hi all, > > This patch series replaces the existing PHYLIB integration with PHYLINK which > is a superior solution since we need to support a collection of fixed links, > integrated PHYs, SFP, non-pluggable SFPs etc. > > I am expecting quite a lot of breakage, for a number of reasons: > > - PHYLINK does not create a PHY device for fixed links (MLO_AN_FIXED), which > means that user-facing port (not DSA, not CPU) need to be explicitly handled > with phylink_mac_ops now > FWIW: We (zenitel) does not use fixed-link on user-port. I gave the patches a spin on our board with lan9303. As expected: no breakage. Egil