On Wed, 20 Apr 2011, Guennadi Liakhovetski wrote: > Thanks to all for clarifications. Since everyone is convinced, that that > idle function in mmc bus.c is appropriate, I restored it and managed to > achieve my goals also without it by adjusting the platform runtime pm and > power domain prototype support. > > But I still need those "cheating" calls to > > pm_runtime_put_noidle(&pdev->dev); > and > pm_runtime_get_noresume(&pdev->dev); > > in the sh_mmcif.c driver. If I use the patch as posted at > > http://article.gmane.org/gmane.linux.ports.sh.devel/10724 > > but without those two calls, and load and unload the driver, while a card > is plugged in, unloading the driver doesn't power down the interface, > because the usage_count == 1 also after the kernel has soft-ejected the > card You haven't explained why you want the interface to be powered down when the driver is unloaded. Normally, unloading a driver is pretty much the exact reverse of loading it -- if the interface wasn't powered down before the driver was loaded originally then it shouldn't be powered down after the driver is unloaded. Conversely, if the interface _was_ powered down before the driver was loaded originally, then making unload the exact reverse of load will naturally leave the interface powered down, with no need for any extra "cheating" calls. > mmc0: card 0001 removed > > With my put_noidle() / get_noresume() hack all cases of modprobe / rmmod, > card insert / eject work correctly. Depending on the subsystem, those calls may not be "cheating" at all. When a subsystem has to deal with a variety of device drivers, some of which may not support runtime PM, a standard trick is for the subsystem to increment the usage_count value before probing. Drivers that aren't runtime-PM-aware will leave the usage_count alone and therefore the device won't ever be runtime-suspended. Drivers that _are_ runtime-PM-aware will know to decrement the usage_count in their probe routine and increment it in their remove routine. However this involves calling pm_runtime_put_noidle() during probe and pm_runtime_get_noresume() during remove -- not during module initialization and module unloading. Alan Stern -- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
