On Tuesday, June 17, 2014 10:37:03 PM Rafael J. Wysocki wrote: > On Tuesday, June 17, 2014 10:26:14 PM Rafael J. Wysocki wrote: > > On Tuesday, June 17, 2014 10:11:32 AM Alan Stern wrote: > > > On Mon, 16 Jun 2014, Rafael J. Wysocki wrote: > > > > > > > > For reasons having nothing to do with Allen's suggested change, I > > > > > wonder if we shouldn't replace this line with something like: > > > > > > > > > > - else if (dev->power.disable_depth == 1 && > > > > > dev->power.is_suspended > > > > > + else if (dev->power.disable > 0 && !dev->power.is_suspended > > > > > && dev->power.runtime_status == RPM_ACTIVE) > > > > > retval = 1; > > > > > > > > > > It seems that I've been bitten by this several times in the past. > > > > > When a device is disabled for runtime PM, and more or less permanently > > > > > stuck in the RPM_ACTIVE state, calls to pm_runtime_resume() or > > > > > pm_runtime_get_sync() shouldn't fail. > > > > > > > > > > For example, suppose some devices of a certain type support runtime > > > > > power management but others don't. We naturally want to call > > > > > pm_runtime_disable() for the ones that don't. But we also want the > > > > > same driver to work for all the devices, which means that > > > > > pm_runtime_get_sync() should return success -- otherwise the driver > > > > > will think that something has gone wrong. > > > > > > > > > > Rafael, what do you think? > > > > > > > > That condition is there specifically to take care of the system suspend > > > > code path. It means that if runtime PM is disabled, but it only has > > > > been > > > > disabled by the system suspend code path, we should treat the device as > > > > "active" (ie. return 1). That won't work after the proposed change. > > > > > > Ah, yes, quite true. Okay, suppose we replace that line with just: > > > > > > + else if (dev->power.disable > 0 > > > > > > > I guess drivers that want to work with devices where runtime PM may be > > > > disabled can just check the return value of rpm_resume() for -EACCES? > > > > > > They could, but it's extra work and it's extremely easy to forget > > > about. I'd prefer not to do things that way. > > > > In that case we need to audit all code that checks the return value of > > __pm_runtime_resume() to verify that it doesn't depend on the current > > behavior in any way. It shouldn't, but still. > > > > Also we probably should drop the -EACCES return value from rpm_resume() in > > the > > same patch, because it specifically only covers the dev->power.disable > 0 > > case > > (which BTW is consistent with the suspend side of things, so I'm totally > > unsure > > about that being the right thing to do to be honest). > > Perhaps it'd be better to rework __pm_runtime_resume() to convert the -EACCES > return value from rpm_resume() into 0 if RPM_GET_PUT is set?
Or do something like this? --- drivers/base/power/runtime.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) Index: linux-pm/drivers/base/power/runtime.c =================================================================== --- linux-pm.orig/drivers/base/power/runtime.c +++ linux-pm/drivers/base/power/runtime.c @@ -608,7 +608,8 @@ static int rpm_resume(struct device *dev repeat: if (dev->power.runtime_error) retval = -EINVAL; - else if (dev->power.disable_depth == 1 && dev->power.is_suspended + else if (((dev->power.disable_depth > 0 && (rpmflags & RPM_GET_PUT)) + || (dev->power.disable_depth == 1 && dev->power.is_suspended)) && dev->power.runtime_status == RPM_ACTIVE) retval = 1; else if (dev->power.disable_depth > 0) -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/