On 01/18/2017 12:15 PM, Tony Lindgren wrote: > * Grygorii Strashko <grygorii.strashko@xxxxxx> [170118 10:05]: >> >> >> On 01/18/2017 10:53 AM, Tony Lindgren wrote: >>> Hi, >>> >>> * Bin Liu <b-liu@xxxxxx> [170118 06:26]: >>>> With this v3, I still get -71 error when a device is plugged to a hub to >>>> musb. It doesn't happen though if the device is already plugged to the hub >>>> before attaching the hub to musb. >>>> >>>> [ 177.579300] usb 1-1: reset high-speed USB device number 2 using musb-hdrc >>>> [ 177.719308] usb 1-1: device descriptor read/64, error -71 >>>> [ 178.350297] usb 1-1.1: new high-speed USB device number 4 using musb-hdrc >>> >>> I think the -71 issue is a different regression. I've verified that v4.8.y >>> does not have this, but v4.9.y does have it. >>> >>> So far the easiest way for me to reproduce the -71 problem is by plugging >>> a USB drive into a connected hub. If I connect the hub with USB drive already >>> plugged into the hub, it does not happen. >>> >>> With the hub connected musb is already active when the USB drive is plugged >> >> This is not exactly try, i think, because cppi can be in inactive state >> because of autosuspend (all calls are paired). > > Musb is active when there's something connected meaning cppi41 is clocked > when a hub is connected. The actual runtime PM implementation happens for > the interconnect target module for both musb and cppi41 in hwmod.c code. > >>> in. So I'm now suspecting this -71 regression is not related to runtime PM >>> changes. Bisect and boot and plug devices time I think unless you have >>> better ideas? >>> >>>> And I still get -115 error flooding with thumb drive. >>>> >>>> [ 236.880068] cppi41-dma-engine 47400000.dma-controller: cppi41_irq pm runtime get: -115 >>>> >>>> I tested with TI AM335x GP EVM. The problems happen on both musb ports. >>> >>> OK. So it's pointless to try to play with the autosuspend timeout. >>> >>> Let's just do a WARN(!pm_runtime_active(cdd->ddev.dev->parent)) there >>> until we have musb_cppi41.c dma calls properly paired and don't have >>> dma requests lingering. >>> >>> Care to try the updated patch below? It won't help with the -71 issue >>> but the $subject issue should be fixed. And you should not see the >>> WARN() trigger with your tests presumably. >>> >> >> Sry, but wouldn't be more simple and safe just to drop PM runtime from this driver, >> as it is part of musb and musb PM state expected to be handled properly by PM runtime now. > > Well we still need PM runtime in cppi41 to initialize it when it gets > probed and idle it when not used even if musb modules are not loaded. > > Unfortunately until musb_cppi41.c dma calls are fixed, we can't do > any sane use counting in cppi41.c without adding timeout handling > to it. > Just thinking, may be cppi41 should not be platform device at all and it might be reasonable to have it as lib with cppi41_init()/cppi41_remove(), so musb SoC glue layer will initialize it, because it provides services to other HW blocks withing musb only. -- regards, -grygorii -- To unsubscribe from this list: send the line "unsubscribe dmaengine" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
