On February 16, 2016 3:06:55 AM PST, Sergei Shtylyov <sergei.shtylyov@xxxxxxxxxxxxxxxxxx> wrote: >Hello. > >On 2/16/2016 10:42 AM, Geert Uytterhoeven wrote: > >>> I have observed what appears to be a regression in the ravb ethernet >driver >>> caused by d5c3d84657db ("net: phy: Avoid polling PHY with >>> PHY_IGNORE_INTERRUPTS"). >>> >>> When booting net-next configured with the ARM64 defconfig on the >Renesas >>> r8a7795/salvator-x I see the following and the ravb is unable to >access the >>> network. With the above mentioned patch reverted I am able to boot >to >>> user-space using nfsroot. >> >> The ravb interrupt is connected to a GPIO controller, which is >> runtime-suspended and thus not serving the interrupt. OK that makes complete sense then, is your patch going to make it into an upcoming v4.5-rc pull request or should we think about having the ravb driver re-assign its phydev->irq to PHY_POLL for the time being? >> >> Cfr. "[PATCH/RFC] gpio: rcar: Add Runtime PM handling for interrupts" >> (http://www.spinics.net/lists/linux-renesas-soc/msg00532.html). >> >> I assume it worked before as the PHY driver polled the PHY instead of >relying >> solely on the interrupt. > > Correct. BTW, I'm going to look at handling AVB_PHY_INT in the ravb >driver, thus removing the need for routing it to the GPIO controller >now that >phylib allows this again... That sounds like a better solution if the PHY interrupt is consistently and uniformly available on these boards. -- Florian