Hi, (please avoid top-posting) On Wed, Jun 25, 2014 at 10:59:57AM +0200, Pedro Erencia wrote: > Hi, > > Thanks for the suggestions Felipe. > > We are not processing our packets in during giveback, but anyway I've > tried your patch and it results in a very little (~5%) increase in > transfer rate. > > I think it is a latency issue, since running the application reports a > 97% idle time on top. > > The profiling of the kernel functions gives the following > > > 30238 total 0.0046 > 22903 omap3_enter_idle 77.3750 hmm... can you disable autoidle on musb and see if it helps. The simplest way is to an unbalanced pm_runtime_get_sync() in the driver, so that usage counter never reaches zero. You can also just try setting a larger timeout: diff --git a/drivers/usb/musb/musb_core.c b/drivers/usb/musb/musb_core.c index eff3c5c..7e28017 100644 --- a/drivers/usb/musb/musb_core.c +++ b/drivers/usb/musb/musb_core.c @@ -1893,7 +1893,7 @@ musb_init_controller(struct device *dev, int nIrq, void __iomem *ctrl) } pm_runtime_use_autosuspend(musb->controller); - pm_runtime_set_autosuspend_delay(musb->controller, 200); + pm_runtime_set_autosuspend_delay(musb->controller, 5000); pm_runtime_enable(musb->controller); spin_lock_init(&musb->lock); -- balbi
Attachment:
signature.asc
Description: Digital signature
