30.09.2019 20:46, Peter Geis пишет: > On Mon, Sep 30, 2019 at 1:42 PM Dmitry Osipenko <digetx@xxxxxxxxx> wrote: >> >> 30.09.2019 18:54, Peter Geis пишет: >>> Dmitry, >>> >>> As far as I can tell the cpuidle drivers work perfectly, but I don't >>> have full power management yet on either my T20 device nor my T30 >>> device. >>> They aren't the cause of this though. >>> >>> I've been sticking to the mainline release code since 5.3 landed, as >>> something was merged into linux-next that breaks brcm4329/brcm4330 >>> firmware loading. >>> >>> Jumping to linux-next to test your driver just revealed the behavior. >>> >>> On my T20 device I haven't encountered issues, but that operates >>> almost exclusively in gadget mode. >>> On my T30 device tegra-udc is misbehaving, especially on linux-next. >>> >>> By removing the hardcoded LL_DEBUG config and moving to a command line >>> earlycon statement, I seem to be making progress in capturing what's >>> going on. >>> With the following actions, I got a panic crash dump: >>> phy set to peripheral, boot with tegra-ehci in host mode, usb hub plugged in. >>> Booted successfully, hub enumerated, passed data through attached >>> ethernet device. >>> Unbind the tegra-ehci driver, bind the tegra-udc driver. >>> Hub enumerates, as well as attached ethernet device, but writes to the >>> device throw constant errors. >> >> Is the host mode working properly when booting with the tegra-udc as a >> primary driver? > > I actually got it working <once>. > It threw constant reset errors though. > Afterwards a soft reboot caused the kernel to crash with an irq-init error. > tegra-ehci is the only stable host driver. Okay, I guess the host mode requires some extra configuration (or some quirk) that is currently missing in the tegra-udc. You may try to compare what tegra-ehci and tegra-udc do differently and then figure out how make it to work, for now I don't have any better advises. >>> Unbind the tegra-udc driver, produces the following panic: >> >> Looks like it is a CI driver bug, perhaps timer's tear down isn't >> performed correctly. Maybe it is related to the errors you're seeing and >> something getting stuck with the offending timer being active during of >> the driver's unbinding.. >> >> Probably Peter Chen could help with that. >> >> [snip]