On Thu, 2016-08-25 at 09:17 -0700, Tony Lindgren wrote: > * Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx> [160825 04:53]: > > > > On Wed, 2016-08-24 at 08:43 -0700, Tony Lindgren wrote: > > > > > > > > > PM runtime with UARTs has been working for me just fine for quite > > > some > > > time now. I'm mostly using omap-serial still and have not checked > > > with > > > 8250-omap recently. > > > > Yeah, they both are using irq_safe flag. Actually they two out of > > only > > 15 users in entire kernel. > > > > Looks like OMAP code is main user of the flag. > > Yeah and we've seen that using irq_safe is not a good solution. > > > > > > > > > But yeah I agree that PM runtime should be improved and simplified > > > for > > > 8250. > > > > Why not to do this once for all in serial_core.c directly? This is an approach I'm developing. See below. > > > > > > > > We should not use irq_safe at all and instead fix up things so > > > runtime_resume wakes up things with pm_runtime_get without > > > get_sync. > > > This adds latency only to the first wake-up event which can > > > already > > > take a long time at least on omaps as things are getting powered > > > back > > > on. > > > > > > > > No idea how to implement that though for cases when there's no > > > dedicated wakeirq.. Maybe change the irq to a threaded irq for the > > > duration of idle for 8250 irq? > > > > I have no ideas either, but irq_safe should be not used if we are > > doing > > pm_runtime_get_sync(). > > Right. How about try something like this to play with: > > 1. Configure an external wakeup source as a wakeirq, could be whatever > GPIO pin you can trigger for testing too > > 2. Then in 8250 pm_runtime_suspend do disable_irq on the 8250 irq > > 3. Trigger a wakeirq event > > 4. In 8250 pm_runtime_resume, enable_irq on the 8250 irq > > And then no irq_safe needed and we can get rid of all the > pm_runtime calls in interrupt handlers :) To all who involved: I have just pushed a proof-of-concept series to my public tree [1] and we may discuss if it's the direction we would go. The series has been tested on our hardware with serial console and RxD used as GPIO to wake it. [1] https://bitbucket.org/andy-shev/linux/branch/topic%2Fuart%2frpm -- Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx> Intel Finland Oy -- To unsubscribe from this list: send the line "unsubscribe linux-serial" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
