| Message ID | 545B9A6D.5060206@samsung.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
[...] >> >> Yes they would, although they require some minor additional adaptations. >> >> Those resources that are enabled from the driver's runtime PM resume >> callback, should also be enabled during ->probe(). The >> pm_runtime_set_active() will then update the state to reflect this. >> >> Then, if CONFIG_PM_RUNTIME is enabled - the device will be scheduled >> to go inactive from driver core (pm_request_idle()), after ->probe() >> has completed. Thus saving power if it's unused. >> If CONFIG_PM_RUNTIME isn't enabled - the driver will still be >> functional, since all resources are enabled during ->probe(). > > OK, I suspected you also assumed enabling relevant resources, so the PM > state matches the hardware state. > > Sorry for getting back late to this, now there is a regression on > Exynos seen in multiple drivers, i.e. affecting all the media and > video devices. Is this patch series going to be merged for v3.18 as > a regression fix ? If so I would need to update remaining drivers to > enable clocks and use pm_runtime_set_active() in probe(). Urgh!!! Let's see how we can work this out. I will be helping out and give this the highest prio! I did post a patchset for exynos 5 media gsc driver, I guess you have seen it!? Now, that doesn't help us much but still. https://www.mail-archive.com/linux-media@vger.kernel.org/msg80592.html > > I can see two options to fix bugs which appeared in Exynos after > merging the patch series switching to the OF generic power domain API: > > 1. merge this series and update the affected drivers for v3.18, This version is superseded by a v3. That takes a more solid approach on how to power on the PM domain from the bus' ->probe(). It's being discussed currently. [PATCH v3 0/9] PM / Domains: Fix race conditions during boot You should be on cc-list of that. > > 2. revert for now to the previous behaviour, doing something as > the patch below. > > 1. seems only a partial solution, since the regression remains for the > loadable modules which are loaded after late_initcall(). At that point > power domain may be disabled and the driver attempting to access > the hardware will hand the system. That was a limitation which is fixed in v3. I have tested loading modules and it works nicely. > > It's also a bit not clear to me why there is an assumption that when > a power domain is initially enabled all its corresponding devices are > already also fully activated ? That's a very good question! I think the assumption is wrong! Somehow we need to decouple that dependency. To me, typically the "need_restore" flag should reflect the current runtime PM status of the device, which isn't the case right now. In the v3 version of this patchset, the PM domain will be powered on from the bus's ->probe(), which also means the "need_restore" flag will initially always be set to "false" once the device are being probed by the driver. That means, those drivers not invoking pm_runtime_set_active() and manually enable clk/resources during ->probe(), but instead relies on pm_runtime_get_sync() - will _not_ work. As you stated above for some of the Exynos drivers. Even if it's likely that most of these Exynos drivers should be using pm_runtime_set_active() during ->probe(), the key-problem lies in genpd's wrong assumption about the device's runtime PM status, which is stored in the "need restore" flag. If we would fix the issue in genpd for the need_restore flag, that should solve the regression for Exynos, don't you think? I will immediately start working on a patch on genpd, which tries this approach, I will keep you posted. > > int __pm_genpd_add_device(struct generic_pm_domain *genpd, struct device *dev, > struct gpd_timing_data *td) > { > ... > gpd_data->need_restore = genpd->status == GPD_STATE_POWER_OFF; > ... > } > > It seems correct to me to have initially the power domain enabled and some > devices inactive, e.g. if device's driver manages its clocks and didn't > turn them on yet. As stated, I fully agree! [...] Kind regards Uffe
diff --git a/arch/arm/mach-exynos/pm_domains.c b/arch/arm/mach-exynos/pm_domains.c index 20f2671..0b0bf68 100644 --- a/arch/arm/mach-exynos/pm_domains.c +++ b/arch/arm/mach-exynos/pm_domains.c @@ -157,7 +157,7 @@ static __init int exynos4_pm_init_power_domain(void) no_clk: on = __raw_readl(pd->base + 0x4) & INT_LOCAL_PWR_EN; - pm_genpd_init(&pd->pd, NULL, !on); + pm_genpd_init(&pd->pd, NULL, !on, true); of_genpd_add_provider_simple(np, &pd->pd); } diff --git a/arch/arm/mach-s3c64xx/pm.c b/arch/arm/mach-s3c64xx/pm.c index aaf7bea..a347654 100644 --- a/arch/arm/mach-s3c64xx/pm.c +++ b/arch/arm/mach-s3c64xx/pm.c @@ -315,10 +315,10 @@ int __init s3c64xx_pm_init(void) for (i = 0; i < ARRAY_SIZE(s3c64xx_always_on_pm_domains); i++) pm_genpd_init(&s3c64xx_always_on_pm_domains[i]->pd, - &pm_domain_always_on_gov, false); + &pm_domain_always_on_gov, false, false); for (i = 0; i < ARRAY_SIZE(s3c64xx_pm_domains); i++) - pm_genpd_init(&s3c64xx_pm_domains[i]->pd, NULL, false); + pm_genpd_init(&s3c64xx_pm_domains[i]->pd, NULL, false, false); #ifdef CONFIG_S3C_DEV_FB if (dev_get_platdata(&s3c_device_fb.dev)) diff --git a/arch/arm/mach-shmobile/pm-r8a7779.c b/arch/arm/mach-shmobile/pm-r8a7779.c index 82fe3d7..fcb17f7 100644 --- a/arch/arm/mach-shmobile/pm-r8a7779.c +++ b/arch/arm/mach-shmobile/pm-r8a7779.c @@ -83,7 +83,7 @@ static void r8a7779_init_pm_domain(struct r8a7779_pm_domain *r8a7779_pd) { struct generic_pm_domain *genpd = &r8a7779_pd->genpd; - pm_genpd_init(genpd, NULL, false); + pm_genpd_init(genpd, NULL, false, false); genpd->dev_ops.stop = pm_clk_suspend; genpd->dev_ops.start = pm_clk_resume; genpd->dev_ops.active_wakeup = pd_active_wakeup; diff --git a/arch/arm/mach-shmobile/pm-rmobile.c b/arch/arm/mach-shmobile/pm-rmobile.c index 717e641..e8c913e 100644 --- a/arch/arm/mach-shmobile/pm-rmobile.c +++ b/arch/arm/mach-shmobile/pm-rmobile.c @@ -106,7 +106,7 @@ static void rmobile_init_pm_domain(struct rmobile_pm_domain *rmobile_pd) struct generic_pm_domain *genpd = &rmobile_pd->genpd; struct dev_power_governor *gov = rmobile_pd->gov; - pm_genpd_init(genpd, gov ? : &simple_qos_governor, false); + pm_genpd_init(genpd, gov ? : &simple_qos_governor, false, false); genpd->dev_ops.stop = pm_clk_suspend; genpd->dev_ops.start = pm_clk_resume; genpd->dev_ops.active_wakeup = rmobile_pd_active_wakeup; diff --git a/drivers/base/power/domain.c b/drivers/base/power/domain.c index 40bc2f4..6a2da45 100644 --- a/drivers/base/power/domain.c +++ b/drivers/base/power/domain.c @@ -1442,7 +1442,8 @@ int __pm_genpd_add_device(struct generic_pm_domain *genpd, struct device *dev, mutex_lock(&gpd_data->lock); gpd_data->base.dev = dev; list_add_tail(&gpd_data->base.list_node, &genpd->dev_list); - gpd_data->need_restore = genpd->status == GPD_STATE_POWER_OFF; + gpd_data->need_restore = (genpd->status == GPD_STATE_POWER_OFF || + genpd->force_restore); gpd_data->td.constraint_changed = true; gpd_data->td.effective_constraint_ns = -1; mutex_unlock(&gpd_data->lock); @@ -1857,9 +1858,11 @@ static int pm_genpd_default_restore_state(struct device *dev) * @genpd: PM domain object to initialize. * @gov: PM domain governor to associate with the domain (may be NULL). * @is_off: Initial value of the domain's power_is_off field. + * @force_restore: True to set device's being added "need_restore" flag. */ void pm_genpd_init(struct generic_pm_domain *genpd, - struct dev_power_governor *gov, bool is_off) + struct dev_power_governor *gov, bool is_off, + bool force_restore) { if (IS_ERR_OR_NULL(genpd)) return; @@ -1873,6 +1876,7 @@ void pm_genpd_init(struct generic_pm_domain *genpd, genpd->in_progress = 0; atomic_set(&genpd->sd_count, 0); genpd->status = is_off ? GPD_STATE_POWER_OFF : GPD_STATE_ACTIVE; + genpd->force_restore = force_restore; init_waitqueue_head(&genpd->status_wait_queue); genpd->poweroff_task = NULL; genpd->resume_count = 0; diff --git a/include/linux/pm_domain.h b/include/linux/pm_domain.h index 73e938b..1acfe05 100644 --- a/include/linux/pm_domain.h +++ b/include/linux/pm_domain.h @@ -71,6 +71,7 @@ struct generic_pm_domain { s64 max_off_time_ns; /* Maximum allowed "suspended" time. */ bool max_off_time_changed; bool cached_power_down_ok; + bool force_restore; /* True to always initially restore devices */ struct gpd_cpuidle_data *cpuidle_data; void (*attach_dev)(struct device *dev); void (*detach_dev)(struct device *dev); @@ -141,7 +142,8 @@ extern int pm_genpd_name_attach_cpuidle(const char *name, int state); extern int pm_genpd_detach_cpuidle(struct generic_pm_domain *genpd); extern int pm_genpd_name_detach_cpuidle(const char *name); extern void pm_genpd_init(struct generic_pm_domain *genpd, - struct dev_power_governor *gov, bool is_off); + struct dev_power_governor *gov, bool is_off, + bool force_restore); extern int pm_genpd_poweron(struct generic_pm_domain *genpd);