Hi Florian, Florian Fainelli <[email protected]> wrote on Thu, 17 Jan 2019 10:00:46 -0800:
> On 1/17/19 7:50 AM, Miquel Raynal wrote: > > Hi Andrew, Vivien, > > > > Vivien Didelot <[email protected]> wrote on Thu, 17 Jan 2019 > > 10:46:41 -0500: > > > >> Hi, > >> > >> On Wed, 16 Jan 2019 23:23:29 +0100, Andrew Lunn <[email protected]> wrote: > >>> Hi Florian > >>> > >>>> A possible approach could be to call the port_disable, port_enable > >>>> callbacks from dsa_slave_suspend() and dsa_slave_resume(), I might have > >>>> some patches doing that already somewhere. > >>> > >>> I expect it is also on Viviens TODO list, since this really could be > >>> in the core. > >> > >> Indeed that is! > > > > So, shall I wait for Vivien's patches (adding port_disable/enable() > > in dsa_slave_suspend/resume()) and keep the driver as-is or do you want > > me to manually call port_disable/enable() from the mv88e6xxx driver? > > Up to you guys, the only thing that I an tell you is that my platform > loses its register contents during suspend/resume, therefore you must > make sure the driver re-applies the entire switch configuration, > identical to how it was prior to suspend. If you need me to test > something, please holler. I am not sure to understand what is lost. On my setup ethtool shows that everything is fine after resume but maybe I fall into a "default" working case. When I compare with the two drivers pointed out by Andrew: * qca8k resume callback: * enable ports, * call dsa_switch_resume(). * bcm_sf2 resume callback: * call dsa_switch_resume(), * reset the switch, * refresh rules (Not applicable?), * enable the PHYs, * setup the ports, * configure vlan (Not applicable?), * mv88e6xxx resume callback: * reset the switch, * enable the PHYs, * setup the ports, * enable IRQs, * call dsa_switch_resume(). This looks pretty similar. Maybe the ports setup are set to default values while I should save some parameters at suspend? I changed a few parameters (like the MTU or the queue length) but they seem to be correct across suspend cycles. Can you show a diff of what part of the configuration is lost? Anyway, I have little background working with switches, so I might miss something big. Thanks, Miquèl
