> We get called twice on suspend, once with DVACT_SUSPEND and once with > DVACT_POWERDOWN. So, here is a patch that does it like in ahci.c an does > everything in the DVACT_POWERDOWN path and nothing in the DVACT_SUSPEND > path.
I suppose so. > @@ -469,6 +512,11 @@ nvme_activate(struct nvme_softc *sc, int act) > rv = config_activate_children(&sc->sc_dev, act); > nvme_shutdown(sc); > break; > + case DVACT_RESUME: > + rv = nvme_resume(sc); > + if (rv == 0) > + rv = config_activate_children(&sc->sc_dev, act); > + break; > default: > rv = config_activate_children(&sc->sc_dev, act); > break; However, note that nvme_resume() cannot fail. If it fails, and you don't resume the children, all sorts of stuff goes wrong. Also your nvme_resume() function performs diagnostic printf's. During a resume, those might not work, fact is they may make the situation worse. Imagine a console screen. Sure it is just printing characters.. unless it has to scroll, now you are running a tremendous amount of code in another driver during a resume function. So the trend here should be to write code which silently just works. Most of the resume code is written this way. Try to get it right in a minimal fashion, and create as few side effects as possible. So perhaps remove the diagnostics checking for failure later. Such diagnostics could make a failure to resume act very strange, leading someone astray quite far.
