| Message ID | 20250606111749.3142348-1-claudiu.beznea.uj@bp.renesas.com |
|---|---|
| Headers | show |
| Series | PM: domains: add devm_pm_domain_attach() | expand |
On Fri, Jun 6, 2025 at 10:01 PM Rafael J. Wysocki <rafael@kernel.org> wrote: > > On Fri, Jun 6, 2025 at 8:55 PM Dmitry Torokhov > <dmitry.torokhov@gmail.com> wrote: > > > > On Fri, Jun 06, 2025 at 06:00:34PM +0200, Rafael J. Wysocki wrote: > > > On Fri, Jun 6, 2025 at 1:18 PM Claudiu <claudiu.beznea@tuxon.dev> wrote: > > > > > > > > From: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com> > > > > > > > > The dev_pm_domain_attach() function is typically used in bus code alongside > > > > dev_pm_domain_detach(), often following patterns like: > > > > > > > > static int bus_probe(struct device *_dev) > > > > { > > > > struct bus_driver *drv = to_bus_driver(dev->driver); > > > > struct bus_device *dev = to_bus_device(_dev); > > > > int ret; > > > > > > > > // ... > > > > > > > > ret = dev_pm_domain_attach(_dev, true); > > > > if (ret) > > > > return ret; > > > > > > > > if (drv->probe) > > > > ret = drv->probe(dev); > > > > > > > > // ... > > > > } > > > > > > > > static void bus_remove(struct device *_dev) > > > > { > > > > struct bus_driver *drv = to_bus_driver(dev->driver); > > > > struct bus_device *dev = to_bus_device(_dev); > > > > > > > > if (drv->remove) > > > > drv->remove(dev); > > > > dev_pm_domain_detach(_dev); > > > > } > > > > > > > > When the driver's probe function uses devres-managed resources that depend > > > > on the power domain state, those resources are released later during > > > > device_unbind_cleanup(). > > > > > > > > Releasing devres-managed resources that depend on the power domain state > > > > after detaching the device from its PM domain can cause failures. > > > > > > > > For example, if the driver uses devm_pm_runtime_enable() in its probe > > > > function, and the device's clocks are managed by the PM domain, then > > > > during removal the runtime PM is disabled in device_unbind_cleanup() after > > > > the clocks have been removed from the PM domain. It may happen that the > > > > devm_pm_runtime_enable() action causes the device to be runtime-resumed. > > > > > > Don't use devm_pm_runtime_enable() then. > > > > What about other devm_ APIs? Are you suggesting that platform drivers > > should not be using devm_clk*(), devm_regulator_*(), > > devm_request_*_irq() and devm_add_action_or_reset()? Because again, > > dev_pm_domain_detach() that is called by platform bus_remove() may shut > > off the device too early, before cleanup code has a chance to execute > > proper cleanup. > > > > The issue is not limited to runtime PM. > > > > > > > > > If the driver specific runtime PM APIs access registers directly, this > > > > will lead to accessing device registers without clocks being enabled. > > > > Similar issues may occur with other devres actions that access device > > > > registers. > > > > > > > > Add devm_pm_domain_attach(). When replacing the dev_pm_domain_attach() and > > > > dev_pm_domain_detach() in bus probe and bus remove, it ensures that the > > > > device is detached from its PM domain in device_unbind_cleanup(), only > > > > after all driver's devres-managed resources have been release. > > > > > > > > For flexibility, the implemented devm_pm_domain_attach() has 2 state > > > > arguments, one for the domain state on attach, one for the domain state on > > > > detach. > > > > > > dev_pm_domain_attach() is not part driver API and I'm not convinced at > > > > Is the concern that devm_pm_domain_attach() will be [ab]used by drivers? > > Yes, among other things. This would be much less objectionable to me if it were not devm_, but also the current expectation is that the PM domain will be gone after device_remove() has returned.
On Fri, 6 Jun 2025 22:01:52 +0200 "Rafael J. Wysocki" <rafael@kernel.org> wrote: Hi Rafael, > On Fri, Jun 6, 2025 at 8:55 PM Dmitry Torokhov > <dmitry.torokhov@gmail.com> wrote: > > > > On Fri, Jun 06, 2025 at 06:00:34PM +0200, Rafael J. Wysocki wrote: > > > On Fri, Jun 6, 2025 at 1:18 PM Claudiu <claudiu.beznea@tuxon.dev> wrote: > > > > > > > > From: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com> > > > > > > > > The dev_pm_domain_attach() function is typically used in bus code alongside > > > > dev_pm_domain_detach(), often following patterns like: > > > > > > > > static int bus_probe(struct device *_dev) > > > > { > > > > struct bus_driver *drv = to_bus_driver(dev->driver); > > > > struct bus_device *dev = to_bus_device(_dev); > > > > int ret; > > > > > > > > // ... > > > > > > > > ret = dev_pm_domain_attach(_dev, true); > > > > if (ret) > > > > return ret; > > > > > > > > if (drv->probe) > > > > ret = drv->probe(dev); > > > > > > > > // ... > > > > } > > > > > > > > static void bus_remove(struct device *_dev) > > > > { > > > > struct bus_driver *drv = to_bus_driver(dev->driver); > > > > struct bus_device *dev = to_bus_device(_dev); > > > > > > > > if (drv->remove) > > > > drv->remove(dev); > > > > dev_pm_domain_detach(_dev); > > > > } > > > > > > > > When the driver's probe function uses devres-managed resources that depend > > > > on the power domain state, those resources are released later during > > > > device_unbind_cleanup(). > > > > > > > > Releasing devres-managed resources that depend on the power domain state > > > > after detaching the device from its PM domain can cause failures. > > > > > > > > For example, if the driver uses devm_pm_runtime_enable() in its probe > > > > function, and the device's clocks are managed by the PM domain, then > > > > during removal the runtime PM is disabled in device_unbind_cleanup() after > > > > the clocks have been removed from the PM domain. It may happen that the > > > > devm_pm_runtime_enable() action causes the device to be runtime-resumed. > > > > > > Don't use devm_pm_runtime_enable() then. > > > > What about other devm_ APIs? Are you suggesting that platform drivers > > should not be using devm_clk*(), devm_regulator_*(), > > devm_request_*_irq() and devm_add_action_or_reset()? Because again, > > dev_pm_domain_detach() that is called by platform bus_remove() may shut > > off the device too early, before cleanup code has a chance to execute > > proper cleanup. > > > > The issue is not limited to runtime PM. > > > > > > > > > If the driver specific runtime PM APIs access registers directly, this > > > > will lead to accessing device registers without clocks being enabled. > > > > Similar issues may occur with other devres actions that access device > > > > registers. > > > > > > > > Add devm_pm_domain_attach(). When replacing the dev_pm_domain_attach() and > > > > dev_pm_domain_detach() in bus probe and bus remove, it ensures that the > > > > device is detached from its PM domain in device_unbind_cleanup(), only > > > > after all driver's devres-managed resources have been release. > > > > > > > > For flexibility, the implemented devm_pm_domain_attach() has 2 state > > > > arguments, one for the domain state on attach, one for the domain state on > > > > detach. > > > > > > dev_pm_domain_attach() is not part driver API and I'm not convinced at > > > > Is the concern that devm_pm_domain_attach() will be [ab]used by drivers? > > Yes, among other things. Maybe naming could make abuse at least obvious to spot? e.g. pm_domain_attach_with_devm_release() > > > In that case we can go back to using devres group to enforce ordering, > > but proper ordering is needed. > > Sure. Ok. Please take a look at: https://lore.kernel.org/all/20250215130849.227812-1-claudiu.beznea.uj@bp.renesas.com/ > > > > all by the arguments above. > > > > Please reconsider given the fact that issue is not limited to the > > runtime PM. > > PM domains are not resources, they are interfaces that are added to > devices by the bus types that need them and they also need to be > removed by those bus types. > > A PM domain needs to go away at remove time because it may not make > sense to use PM domain callbacks without driver callbacks and if > enabled runtime PM is leaked beyond the point at which there are no > driver and bus type callbacks, this is exactly what may happen. I'm fully in agreement with all that. However, I'm not sure on relevance to this discussion as (if the new function is used as intended) the pm domain will get removed before any problems occur. The only call in __device_release_driver() between the bus remove and device_unbind_cleanup() which unrolls the devm stuff as the first thing it does is dev->bus->dma_cleanup(). Maybe I'm missing another path? > > If you have ordering issues in drivers, that's where they are and > that's where they need to be addressed. > To my viewpoint, the ordering issue is in the bus driver because devm setup calls are in device driver probe() which comes after the bus_type->probe() but unwound after the bus_type->remove() - they should be before that if they are only for device driver usage. There are devm uses in bus probe functions though which is assume why the ordering is as we have it. Personally I preferred the devres group in the bus driver. It's a model a couple of busses are already using. The solution in this patch also worked for me though as a good compromise between different view points. Claudiu has been working on different solutions to this problem for a long time so lets work together to find a solution that finally resolves it. Jonathan > Thanks!
On Sat, Jun 7, 2025 at 3:06 PM Jonathan Cameron <jic23@kernel.org> wrote: > > On Fri, 6 Jun 2025 22:01:52 +0200 > "Rafael J. Wysocki" <rafael@kernel.org> wrote: > > Hi Rafael, > > > On Fri, Jun 6, 2025 at 8:55 PM Dmitry Torokhov > > <dmitry.torokhov@gmail.com> wrote: > > > > > > On Fri, Jun 06, 2025 at 06:00:34PM +0200, Rafael J. Wysocki wrote: > > > > On Fri, Jun 6, 2025 at 1:18 PM Claudiu <claudiu.beznea@tuxon.dev> wrote: > > > > > > > > > > From: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com> > > > > > > > > > > The dev_pm_domain_attach() function is typically used in bus code alongside > > > > > dev_pm_domain_detach(), often following patterns like: > > > > > > > > > > static int bus_probe(struct device *_dev) > > > > > { > > > > > struct bus_driver *drv = to_bus_driver(dev->driver); > > > > > struct bus_device *dev = to_bus_device(_dev); > > > > > int ret; > > > > > > > > > > // ... > > > > > > > > > > ret = dev_pm_domain_attach(_dev, true); > > > > > if (ret) > > > > > return ret; > > > > > > > > > > if (drv->probe) > > > > > ret = drv->probe(dev); > > > > > > > > > > // ... > > > > > } > > > > > > > > > > static void bus_remove(struct device *_dev) > > > > > { > > > > > struct bus_driver *drv = to_bus_driver(dev->driver); > > > > > struct bus_device *dev = to_bus_device(_dev); > > > > > > > > > > if (drv->remove) > > > > > drv->remove(dev); > > > > > dev_pm_domain_detach(_dev); > > > > > } > > > > > > > > > > When the driver's probe function uses devres-managed resources that depend > > > > > on the power domain state, those resources are released later during > > > > > device_unbind_cleanup(). > > > > > > > > > > Releasing devres-managed resources that depend on the power domain state > > > > > after detaching the device from its PM domain can cause failures. > > > > > > > > > > For example, if the driver uses devm_pm_runtime_enable() in its probe > > > > > function, and the device's clocks are managed by the PM domain, then > > > > > during removal the runtime PM is disabled in device_unbind_cleanup() after > > > > > the clocks have been removed from the PM domain. It may happen that the > > > > > devm_pm_runtime_enable() action causes the device to be runtime-resumed. > > > > > > > > Don't use devm_pm_runtime_enable() then. > > > > > > What about other devm_ APIs? Are you suggesting that platform drivers > > > should not be using devm_clk*(), devm_regulator_*(), > > > devm_request_*_irq() and devm_add_action_or_reset()? Because again, > > > dev_pm_domain_detach() that is called by platform bus_remove() may shut > > > off the device too early, before cleanup code has a chance to execute > > > proper cleanup. > > > > > > The issue is not limited to runtime PM. > > > > > > > > > > > > If the driver specific runtime PM APIs access registers directly, this > > > > > will lead to accessing device registers without clocks being enabled. > > > > > Similar issues may occur with other devres actions that access device > > > > > registers. > > > > > > > > > > Add devm_pm_domain_attach(). When replacing the dev_pm_domain_attach() and > > > > > dev_pm_domain_detach() in bus probe and bus remove, it ensures that the > > > > > device is detached from its PM domain in device_unbind_cleanup(), only > > > > > after all driver's devres-managed resources have been release. > > > > > > > > > > For flexibility, the implemented devm_pm_domain_attach() has 2 state > > > > > arguments, one for the domain state on attach, one for the domain state on > > > > > detach. > > > > > > > > dev_pm_domain_attach() is not part driver API and I'm not convinced at > > > > > > Is the concern that devm_pm_domain_attach() will be [ab]used by drivers? > > > > Yes, among other things. > > Maybe naming could make abuse at least obvious to spot? e.g. > pm_domain_attach_with_devm_release() If I'm not mistaken, it is not even necessary to use devres for this. You might as well add a dev_pm_domain_detach() call to device_unbind_cleanup() after devres_release_all(). There is a slight complication related to the second argument of it, but I suppose that this can be determined at the attach time and stored in a new device PM flag, or similar. Note that dev->pm_domain is expected to be cleared by ->detach(), so this should not cause the domain to be detached twice in a row from the same device, but that needs to be double-checked. Thanks!
Hi, Rafael, On 09.06.2025 22:59, Rafael J. Wysocki wrote: > On Sat, Jun 7, 2025 at 3:06 PM Jonathan Cameron <jic23@kernel.org> wrote: >> >> On Fri, 6 Jun 2025 22:01:52 +0200 >> "Rafael J. Wysocki" <rafael@kernel.org> wrote: >> >> Hi Rafael, >> >>> On Fri, Jun 6, 2025 at 8:55 PM Dmitry Torokhov >>> <dmitry.torokhov@gmail.com> wrote: >>>> >>>> On Fri, Jun 06, 2025 at 06:00:34PM +0200, Rafael J. Wysocki wrote: >>>>> On Fri, Jun 6, 2025 at 1:18 PM Claudiu <claudiu.beznea@tuxon.dev> wrote: >>>>>> >>>>>> From: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com> >>>>>> >>>>>> The dev_pm_domain_attach() function is typically used in bus code alongside >>>>>> dev_pm_domain_detach(), often following patterns like: >>>>>> >>>>>> static int bus_probe(struct device *_dev) >>>>>> { >>>>>> struct bus_driver *drv = to_bus_driver(dev->driver); >>>>>> struct bus_device *dev = to_bus_device(_dev); >>>>>> int ret; >>>>>> >>>>>> // ... >>>>>> >>>>>> ret = dev_pm_domain_attach(_dev, true); >>>>>> if (ret) >>>>>> return ret; >>>>>> >>>>>> if (drv->probe) >>>>>> ret = drv->probe(dev); >>>>>> >>>>>> // ... >>>>>> } >>>>>> >>>>>> static void bus_remove(struct device *_dev) >>>>>> { >>>>>> struct bus_driver *drv = to_bus_driver(dev->driver); >>>>>> struct bus_device *dev = to_bus_device(_dev); >>>>>> >>>>>> if (drv->remove) >>>>>> drv->remove(dev); >>>>>> dev_pm_domain_detach(_dev); >>>>>> } >>>>>> >>>>>> When the driver's probe function uses devres-managed resources that depend >>>>>> on the power domain state, those resources are released later during >>>>>> device_unbind_cleanup(). >>>>>> >>>>>> Releasing devres-managed resources that depend on the power domain state >>>>>> after detaching the device from its PM domain can cause failures. >>>>>> >>>>>> For example, if the driver uses devm_pm_runtime_enable() in its probe >>>>>> function, and the device's clocks are managed by the PM domain, then >>>>>> during removal the runtime PM is disabled in device_unbind_cleanup() after >>>>>> the clocks have been removed from the PM domain. It may happen that the >>>>>> devm_pm_runtime_enable() action causes the device to be runtime-resumed. >>>>> >>>>> Don't use devm_pm_runtime_enable() then. >>>> >>>> What about other devm_ APIs? Are you suggesting that platform drivers >>>> should not be using devm_clk*(), devm_regulator_*(), >>>> devm_request_*_irq() and devm_add_action_or_reset()? Because again, >>>> dev_pm_domain_detach() that is called by platform bus_remove() may shut >>>> off the device too early, before cleanup code has a chance to execute >>>> proper cleanup. >>>> >>>> The issue is not limited to runtime PM. >>>> >>>>> >>>>>> If the driver specific runtime PM APIs access registers directly, this >>>>>> will lead to accessing device registers without clocks being enabled. >>>>>> Similar issues may occur with other devres actions that access device >>>>>> registers. >>>>>> >>>>>> Add devm_pm_domain_attach(). When replacing the dev_pm_domain_attach() and >>>>>> dev_pm_domain_detach() in bus probe and bus remove, it ensures that the >>>>>> device is detached from its PM domain in device_unbind_cleanup(), only >>>>>> after all driver's devres-managed resources have been release. >>>>>> >>>>>> For flexibility, the implemented devm_pm_domain_attach() has 2 state >>>>>> arguments, one for the domain state on attach, one for the domain state on >>>>>> detach. >>>>> >>>>> dev_pm_domain_attach() is not part driver API and I'm not convinced at >>>> >>>> Is the concern that devm_pm_domain_attach() will be [ab]used by drivers? >>> >>> Yes, among other things. >> >> Maybe naming could make abuse at least obvious to spot? e.g. >> pm_domain_attach_with_devm_release() > > If I'm not mistaken, it is not even necessary to use devres for this. > > You might as well add a dev_pm_domain_detach() call to > device_unbind_cleanup() after devres_release_all(). There is a slight > complication related to the second argument of it, but I suppose that > this can be determined at the attach time and stored in a new device > PM flag, or similar. I can try this as well. Another option I see at the moment would be keep the code added in drivers/base/power/common.c in drivers/base/platform.c, something like: diff --git a/drivers/base/platform.c b/drivers/base/platform.c index 075ec1d1b73a..391d725cd4c7 100644 --- a/drivers/base/platform.c +++ b/drivers/base/platform.c @@ -1376,10 +1376,18 @@ static int platform_uevent(const struct device *dev, struct kobj_uevent_env *env return 0; } +static void platform_dev_pm_domain_detach(struct device *dev, void *res) +{ + bool *power_off = res; + + dev_pm_domain_detach(dev, *power_off); +} + static int platform_probe(struct device *_dev) { struct platform_driver *drv = to_platform_driver(_dev->driver); struct platform_device *dev = to_platform_device(_dev); + bool *power_off; int ret; /* @@ -1396,15 +1404,22 @@ static int platform_probe(struct device *_dev) if (ret < 0) return ret; + power_off = devres_alloc(platform_dev_pm_domain_detach, sizeof(*power_off), + GFP_KERNEL); + if (!power_off) + return -ENOMEM; + ret = dev_pm_domain_attach(_dev, true); - if (ret) + if (ret) { + devres_free(power_off); goto out; + } - if (drv->probe) { + *power_off = true; + devres_add(_dev, power_off); + + if (drv->probe) ret = drv->probe(dev); - if (ret) - dev_pm_domain_detach(_dev, true); - } out: if (drv->prevent_deferred_probe && ret == -EPROBE_DEFER) { @@ -1422,7 +1437,6 @@ static void platform_remove(struct device *_dev) if (drv->remove) drv->remove(dev); - dev_pm_domain_detach(_dev, true); } but this would involve duplicating code, as, sooner or later, this would have to be done for other busses as well. Could you please let me know what option would you prefer so that I can go forward with it? Thank you for your review, Claudiu > > Note that dev->pm_domain is expected to be cleared by ->detach(), so > this should not cause the domain to be detached twice in a row from > the same device, but that needs to be double-checked. > > Thanks!
On Mon, 9 Jun 2025 21:59:57 +0200 "Rafael J. Wysocki" <rafael@kernel.org> wrote: > On Sat, Jun 7, 2025 at 3:06 PM Jonathan Cameron <jic23@kernel.org> wrote: > > > > On Fri, 6 Jun 2025 22:01:52 +0200 > > "Rafael J. Wysocki" <rafael@kernel.org> wrote: > > > > Hi Rafael, > > > > > On Fri, Jun 6, 2025 at 8:55 PM Dmitry Torokhov > > > <dmitry.torokhov@gmail.com> wrote: > > > > > > > > On Fri, Jun 06, 2025 at 06:00:34PM +0200, Rafael J. Wysocki wrote: > > > > > On Fri, Jun 6, 2025 at 1:18 PM Claudiu <claudiu.beznea@tuxon.dev> wrote: > > > > > > > > > > > > From: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com> > > > > > > > > > > > > The dev_pm_domain_attach() function is typically used in bus code alongside > > > > > > dev_pm_domain_detach(), often following patterns like: > > > > > > > > > > > > static int bus_probe(struct device *_dev) > > > > > > { > > > > > > struct bus_driver *drv = to_bus_driver(dev->driver); > > > > > > struct bus_device *dev = to_bus_device(_dev); > > > > > > int ret; > > > > > > > > > > > > // ... > > > > > > > > > > > > ret = dev_pm_domain_attach(_dev, true); > > > > > > if (ret) > > > > > > return ret; > > > > > > > > > > > > if (drv->probe) > > > > > > ret = drv->probe(dev); > > > > > > > > > > > > // ... > > > > > > } > > > > > > > > > > > > static void bus_remove(struct device *_dev) > > > > > > { > > > > > > struct bus_driver *drv = to_bus_driver(dev->driver); > > > > > > struct bus_device *dev = to_bus_device(_dev); > > > > > > > > > > > > if (drv->remove) > > > > > > drv->remove(dev); > > > > > > dev_pm_domain_detach(_dev); > > > > > > } > > > > > > > > > > > > When the driver's probe function uses devres-managed resources that depend > > > > > > on the power domain state, those resources are released later during > > > > > > device_unbind_cleanup(). > > > > > > > > > > > > Releasing devres-managed resources that depend on the power domain state > > > > > > after detaching the device from its PM domain can cause failures. > > > > > > > > > > > > For example, if the driver uses devm_pm_runtime_enable() in its probe > > > > > > function, and the device's clocks are managed by the PM domain, then > > > > > > during removal the runtime PM is disabled in device_unbind_cleanup() after > > > > > > the clocks have been removed from the PM domain. It may happen that the > > > > > > devm_pm_runtime_enable() action causes the device to be runtime-resumed. > > > > > > > > > > Don't use devm_pm_runtime_enable() then. > > > > > > > > What about other devm_ APIs? Are you suggesting that platform drivers > > > > should not be using devm_clk*(), devm_regulator_*(), > > > > devm_request_*_irq() and devm_add_action_or_reset()? Because again, > > > > dev_pm_domain_detach() that is called by platform bus_remove() may shut > > > > off the device too early, before cleanup code has a chance to execute > > > > proper cleanup. > > > > > > > > The issue is not limited to runtime PM. > > > > > > > > > > > > > > > If the driver specific runtime PM APIs access registers directly, this > > > > > > will lead to accessing device registers without clocks being enabled. > > > > > > Similar issues may occur with other devres actions that access device > > > > > > registers. > > > > > > > > > > > > Add devm_pm_domain_attach(). When replacing the dev_pm_domain_attach() and > > > > > > dev_pm_domain_detach() in bus probe and bus remove, it ensures that the > > > > > > device is detached from its PM domain in device_unbind_cleanup(), only > > > > > > after all driver's devres-managed resources have been release. > > > > > > > > > > > > For flexibility, the implemented devm_pm_domain_attach() has 2 state > > > > > > arguments, one for the domain state on attach, one for the domain state on > > > > > > detach. > > > > > > > > > > dev_pm_domain_attach() is not part driver API and I'm not convinced at > > > > > > > > Is the concern that devm_pm_domain_attach() will be [ab]used by drivers? > > > > > > Yes, among other things. > > > > Maybe naming could make abuse at least obvious to spot? e.g. > > pm_domain_attach_with_devm_release() > > If I'm not mistaken, it is not even necessary to use devres for this. > > You might as well add a dev_pm_domain_detach() call to > device_unbind_cleanup() after devres_release_all(). There is a slight > complication related to the second argument of it, but I suppose that > this can be determined at the attach time and stored in a new device > PM flag, or similar. That options sounds good to me. I think this moves dev_pm_domain_detach() call into the the driver core / perhaps device_unbind_cleanup(). It's a noop if a domain was never attached so that should be fine. Given that second parameter, I guess we can't move the dev_pm_domain_attach() into the driver core as well so it is a little odd wrt to balance, but with some documentation that is probably fine. I don't think we really want a bus->remove_after_devres() callback for just this. Ulf what do you think of this approach? Jonathan > > Note that dev->pm_domain is expected to be cleared by ->detach(), so > this should not cause the domain to be detached twice in a row from > the same device, but that needs to be double-checked. > > Thanks! >
On Wed, 11 Jun 2025 12:11:08 +0300 Claudiu Beznea <claudiu.beznea@tuxon.dev> wrote: > Hi, Rafael, > > On 09.06.2025 22:59, Rafael J. Wysocki wrote: > > On Sat, Jun 7, 2025 at 3:06 PM Jonathan Cameron <jic23@kernel.org> wrote: > >> > >> On Fri, 6 Jun 2025 22:01:52 +0200 > >> "Rafael J. Wysocki" <rafael@kernel.org> wrote: > >> > >> Hi Rafael, > >> > >>> On Fri, Jun 6, 2025 at 8:55 PM Dmitry Torokhov > >>> <dmitry.torokhov@gmail.com> wrote: > >>>> > >>>> On Fri, Jun 06, 2025 at 06:00:34PM +0200, Rafael J. Wysocki wrote: > >>>>> On Fri, Jun 6, 2025 at 1:18 PM Claudiu <claudiu.beznea@tuxon.dev> wrote: > >>>>>> > >>>>>> From: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com> > >>>>>> > >>>>>> The dev_pm_domain_attach() function is typically used in bus code alongside > >>>>>> dev_pm_domain_detach(), often following patterns like: > >>>>>> > >>>>>> static int bus_probe(struct device *_dev) > >>>>>> { > >>>>>> struct bus_driver *drv = to_bus_driver(dev->driver); > >>>>>> struct bus_device *dev = to_bus_device(_dev); > >>>>>> int ret; > >>>>>> > >>>>>> // ... > >>>>>> > >>>>>> ret = dev_pm_domain_attach(_dev, true); > >>>>>> if (ret) > >>>>>> return ret; > >>>>>> > >>>>>> if (drv->probe) > >>>>>> ret = drv->probe(dev); > >>>>>> > >>>>>> // ... > >>>>>> } > >>>>>> > >>>>>> static void bus_remove(struct device *_dev) > >>>>>> { > >>>>>> struct bus_driver *drv = to_bus_driver(dev->driver); > >>>>>> struct bus_device *dev = to_bus_device(_dev); > >>>>>> > >>>>>> if (drv->remove) > >>>>>> drv->remove(dev); > >>>>>> dev_pm_domain_detach(_dev); > >>>>>> } > >>>>>> > >>>>>> When the driver's probe function uses devres-managed resources that depend > >>>>>> on the power domain state, those resources are released later during > >>>>>> device_unbind_cleanup(). > >>>>>> > >>>>>> Releasing devres-managed resources that depend on the power domain state > >>>>>> after detaching the device from its PM domain can cause failures. > >>>>>> > >>>>>> For example, if the driver uses devm_pm_runtime_enable() in its probe > >>>>>> function, and the device's clocks are managed by the PM domain, then > >>>>>> during removal the runtime PM is disabled in device_unbind_cleanup() after > >>>>>> the clocks have been removed from the PM domain. It may happen that the > >>>>>> devm_pm_runtime_enable() action causes the device to be runtime-resumed. > >>>>> > >>>>> Don't use devm_pm_runtime_enable() then. > >>>> > >>>> What about other devm_ APIs? Are you suggesting that platform drivers > >>>> should not be using devm_clk*(), devm_regulator_*(), > >>>> devm_request_*_irq() and devm_add_action_or_reset()? Because again, > >>>> dev_pm_domain_detach() that is called by platform bus_remove() may shut > >>>> off the device too early, before cleanup code has a chance to execute > >>>> proper cleanup. > >>>> > >>>> The issue is not limited to runtime PM. > >>>> > >>>>> > >>>>>> If the driver specific runtime PM APIs access registers directly, this > >>>>>> will lead to accessing device registers without clocks being enabled. > >>>>>> Similar issues may occur with other devres actions that access device > >>>>>> registers. > >>>>>> > >>>>>> Add devm_pm_domain_attach(). When replacing the dev_pm_domain_attach() and > >>>>>> dev_pm_domain_detach() in bus probe and bus remove, it ensures that the > >>>>>> device is detached from its PM domain in device_unbind_cleanup(), only > >>>>>> after all driver's devres-managed resources have been release. > >>>>>> > >>>>>> For flexibility, the implemented devm_pm_domain_attach() has 2 state > >>>>>> arguments, one for the domain state on attach, one for the domain state on > >>>>>> detach. > >>>>> > >>>>> dev_pm_domain_attach() is not part driver API and I'm not convinced at > >>>> > >>>> Is the concern that devm_pm_domain_attach() will be [ab]used by drivers? > >>> > >>> Yes, among other things. > >> > >> Maybe naming could make abuse at least obvious to spot? e.g. > >> pm_domain_attach_with_devm_release() > > > > If I'm not mistaken, it is not even necessary to use devres for this. > > > > You might as well add a dev_pm_domain_detach() call to > > device_unbind_cleanup() after devres_release_all(). There is a slight > > complication related to the second argument of it, but I suppose that > > this can be determined at the attach time and stored in a new device > > PM flag, or similar. > > I can try this as well. > > Another option I see at the moment would be keep the code added in > drivers/base/power/common.c in drivers/base/platform.c, something like: Whilst this avoids the exposure of a devm interface by just pushing it down to each bus, we'll for ever be rejecting cleanups that unify it. So I'd rather explore Rafael's suggestion to just handle this one in the driver core. That only deals with the pm domain issue and not the other ones Dmitry refers to where devres being handled after bus->remove() bites us, but maybe that is the right way forwards for now at least. Jonathan > > diff --git a/drivers/base/platform.c b/drivers/base/platform.c > index 075ec1d1b73a..391d725cd4c7 100644 > --- a/drivers/base/platform.c > +++ b/drivers/base/platform.c > @@ -1376,10 +1376,18 @@ static int platform_uevent(const struct device > *dev, struct kobj_uevent_env *env > return 0; > } > > +static void platform_dev_pm_domain_detach(struct device *dev, void *res) > +{ > + bool *power_off = res; > + > + dev_pm_domain_detach(dev, *power_off); > +} > + > static int platform_probe(struct device *_dev) > { > struct platform_driver *drv = to_platform_driver(_dev->driver); > struct platform_device *dev = to_platform_device(_dev); > + bool *power_off; > int ret; > > /* > @@ -1396,15 +1404,22 @@ static int platform_probe(struct device *_dev) > if (ret < 0) > return ret; > > + power_off = devres_alloc(platform_dev_pm_domain_detach, > sizeof(*power_off), > + GFP_KERNEL); > + if (!power_off) > + return -ENOMEM; > + > ret = dev_pm_domain_attach(_dev, true); > - if (ret) > + if (ret) { > + devres_free(power_off); > goto out; > + } > > - if (drv->probe) { > + *power_off = true; > + devres_add(_dev, power_off); > + > + if (drv->probe) > ret = drv->probe(dev); > - if (ret) > - dev_pm_domain_detach(_dev, true); > - } > > out: > if (drv->prevent_deferred_probe && ret == -EPROBE_DEFER) { > @@ -1422,7 +1437,6 @@ static void platform_remove(struct device *_dev) > > if (drv->remove) > drv->remove(dev); > - dev_pm_domain_detach(_dev, true); > } > > but this would involve duplicating code, as, sooner or later, this would > have to be done for other busses as well. > > Could you please let me know what option would you prefer so that I can go > forward with it? > > Thank you for your review, > Claudiu > > > > > Note that dev->pm_domain is expected to be cleared by ->detach(), so > > this should not cause the domain to be detached twice in a row from > > the same device, but that needs to be double-checked. > > > > Thanks! >
On Wed, Jun 11, 2025 at 05:23:07PM +0100, Jonathan Cameron wrote: > On Mon, 9 Jun 2025 21:59:57 +0200 > "Rafael J. Wysocki" <rafael@kernel.org> wrote: > > > On Sat, Jun 7, 2025 at 3:06 PM Jonathan Cameron <jic23@kernel.org> wrote: > > > > > > On Fri, 6 Jun 2025 22:01:52 +0200 > > > "Rafael J. Wysocki" <rafael@kernel.org> wrote: > > > > > > Hi Rafael, > > > > > > > On Fri, Jun 6, 2025 at 8:55 PM Dmitry Torokhov > > > > <dmitry.torokhov@gmail.com> wrote: > > > > > > > > > > On Fri, Jun 06, 2025 at 06:00:34PM +0200, Rafael J. Wysocki wrote: > > > > > > On Fri, Jun 6, 2025 at 1:18 PM Claudiu <claudiu.beznea@tuxon.dev> wrote: > > > > > > > > > > > > > > From: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com> > > > > > > > > > > > > > > The dev_pm_domain_attach() function is typically used in bus code alongside > > > > > > > dev_pm_domain_detach(), often following patterns like: > > > > > > > > > > > > > > static int bus_probe(struct device *_dev) > > > > > > > { > > > > > > > struct bus_driver *drv = to_bus_driver(dev->driver); > > > > > > > struct bus_device *dev = to_bus_device(_dev); > > > > > > > int ret; > > > > > > > > > > > > > > // ... > > > > > > > > > > > > > > ret = dev_pm_domain_attach(_dev, true); > > > > > > > if (ret) > > > > > > > return ret; > > > > > > > > > > > > > > if (drv->probe) > > > > > > > ret = drv->probe(dev); > > > > > > > > > > > > > > // ... > > > > > > > } > > > > > > > > > > > > > > static void bus_remove(struct device *_dev) > > > > > > > { > > > > > > > struct bus_driver *drv = to_bus_driver(dev->driver); > > > > > > > struct bus_device *dev = to_bus_device(_dev); > > > > > > > > > > > > > > if (drv->remove) > > > > > > > drv->remove(dev); > > > > > > > dev_pm_domain_detach(_dev); > > > > > > > } > > > > > > > > > > > > > > When the driver's probe function uses devres-managed resources that depend > > > > > > > on the power domain state, those resources are released later during > > > > > > > device_unbind_cleanup(). > > > > > > > > > > > > > > Releasing devres-managed resources that depend on the power domain state > > > > > > > after detaching the device from its PM domain can cause failures. > > > > > > > > > > > > > > For example, if the driver uses devm_pm_runtime_enable() in its probe > > > > > > > function, and the device's clocks are managed by the PM domain, then > > > > > > > during removal the runtime PM is disabled in device_unbind_cleanup() after > > > > > > > the clocks have been removed from the PM domain. It may happen that the > > > > > > > devm_pm_runtime_enable() action causes the device to be runtime-resumed. > > > > > > > > > > > > Don't use devm_pm_runtime_enable() then. > > > > > > > > > > What about other devm_ APIs? Are you suggesting that platform drivers > > > > > should not be using devm_clk*(), devm_regulator_*(), > > > > > devm_request_*_irq() and devm_add_action_or_reset()? Because again, > > > > > dev_pm_domain_detach() that is called by platform bus_remove() may shut > > > > > off the device too early, before cleanup code has a chance to execute > > > > > proper cleanup. > > > > > > > > > > The issue is not limited to runtime PM. > > > > > > > > > > > > > > > > > > If the driver specific runtime PM APIs access registers directly, this > > > > > > > will lead to accessing device registers without clocks being enabled. > > > > > > > Similar issues may occur with other devres actions that access device > > > > > > > registers. > > > > > > > > > > > > > > Add devm_pm_domain_attach(). When replacing the dev_pm_domain_attach() and > > > > > > > dev_pm_domain_detach() in bus probe and bus remove, it ensures that the > > > > > > > device is detached from its PM domain in device_unbind_cleanup(), only > > > > > > > after all driver's devres-managed resources have been release. > > > > > > > > > > > > > > For flexibility, the implemented devm_pm_domain_attach() has 2 state > > > > > > > arguments, one for the domain state on attach, one for the domain state on > > > > > > > detach. > > > > > > > > > > > > dev_pm_domain_attach() is not part driver API and I'm not convinced at > > > > > > > > > > Is the concern that devm_pm_domain_attach() will be [ab]used by drivers? > > > > > > > > Yes, among other things. > > > > > > Maybe naming could make abuse at least obvious to spot? e.g. > > > pm_domain_attach_with_devm_release() > > > > If I'm not mistaken, it is not even necessary to use devres for this. > > > > You might as well add a dev_pm_domain_detach() call to > > device_unbind_cleanup() after devres_release_all(). There is a slight > > complication related to the second argument of it, but I suppose that > > this can be determined at the attach time and stored in a new device > > PM flag, or similar. > > That options sounds good to me. I think this moves dev_pm_domain_detach() > call into the the driver core / perhaps device_unbind_cleanup(). It's a noop > if a domain was never attached so that should be fine. > > Given that second parameter, I guess we can't move the dev_pm_domain_attach() > into the driver core as well so it is a little odd wrt to balance, > but with some documentation that is probably fine. It is going to be confusing IMO and might lead to ordering issues again if there are more resources allocated by bus probe() before domain attach is called. I know Rafael does not consider dev_pm_domain_attach() a "driver" API (although I do not see much difference between driver and bus probe code), but maybe we can solve this by using different name (devres_controlled_pm_domain_attach() instead of devm_pm_domain_attach()?). OTOH we have devm_pm_domain_attach_list() already which is I guess driver-level API... Thanks.
Hi, Rafael, On 09.06.2025 22:59, Rafael J. Wysocki wrote: > On Sat, Jun 7, 2025 at 3:06 PM Jonathan Cameron <jic23@kernel.org> wrote: >> >> On Fri, 6 Jun 2025 22:01:52 +0200 >> "Rafael J. Wysocki" <rafael@kernel.org> wrote: >> >> Hi Rafael, >> >>> On Fri, Jun 6, 2025 at 8:55 PM Dmitry Torokhov >>> <dmitry.torokhov@gmail.com> wrote: >>>> >>>> On Fri, Jun 06, 2025 at 06:00:34PM +0200, Rafael J. Wysocki wrote: >>>>> On Fri, Jun 6, 2025 at 1:18 PM Claudiu <claudiu.beznea@tuxon.dev> wrote: >>>>>> >>>>>> From: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com> >>>>>> >>>>>> The dev_pm_domain_attach() function is typically used in bus code alongside >>>>>> dev_pm_domain_detach(), often following patterns like: >>>>>> >>>>>> static int bus_probe(struct device *_dev) >>>>>> { >>>>>> struct bus_driver *drv = to_bus_driver(dev->driver); >>>>>> struct bus_device *dev = to_bus_device(_dev); >>>>>> int ret; >>>>>> >>>>>> // ... >>>>>> >>>>>> ret = dev_pm_domain_attach(_dev, true); >>>>>> if (ret) >>>>>> return ret; >>>>>> >>>>>> if (drv->probe) >>>>>> ret = drv->probe(dev); >>>>>> >>>>>> // ... >>>>>> } >>>>>> >>>>>> static void bus_remove(struct device *_dev) >>>>>> { >>>>>> struct bus_driver *drv = to_bus_driver(dev->driver); >>>>>> struct bus_device *dev = to_bus_device(_dev); >>>>>> >>>>>> if (drv->remove) >>>>>> drv->remove(dev); >>>>>> dev_pm_domain_detach(_dev); >>>>>> } >>>>>> >>>>>> When the driver's probe function uses devres-managed resources that depend >>>>>> on the power domain state, those resources are released later during >>>>>> device_unbind_cleanup(). >>>>>> >>>>>> Releasing devres-managed resources that depend on the power domain state >>>>>> after detaching the device from its PM domain can cause failures. >>>>>> >>>>>> For example, if the driver uses devm_pm_runtime_enable() in its probe >>>>>> function, and the device's clocks are managed by the PM domain, then >>>>>> during removal the runtime PM is disabled in device_unbind_cleanup() after >>>>>> the clocks have been removed from the PM domain. It may happen that the >>>>>> devm_pm_runtime_enable() action causes the device to be runtime-resumed. >>>>> >>>>> Don't use devm_pm_runtime_enable() then. >>>> >>>> What about other devm_ APIs? Are you suggesting that platform drivers >>>> should not be using devm_clk*(), devm_regulator_*(), >>>> devm_request_*_irq() and devm_add_action_or_reset()? Because again, >>>> dev_pm_domain_detach() that is called by platform bus_remove() may shut >>>> off the device too early, before cleanup code has a chance to execute >>>> proper cleanup. >>>> >>>> The issue is not limited to runtime PM. >>>> >>>>> >>>>>> If the driver specific runtime PM APIs access registers directly, this >>>>>> will lead to accessing device registers without clocks being enabled. >>>>>> Similar issues may occur with other devres actions that access device >>>>>> registers. >>>>>> >>>>>> Add devm_pm_domain_attach(). When replacing the dev_pm_domain_attach() and >>>>>> dev_pm_domain_detach() in bus probe and bus remove, it ensures that the >>>>>> device is detached from its PM domain in device_unbind_cleanup(), only >>>>>> after all driver's devres-managed resources have been release. >>>>>> >>>>>> For flexibility, the implemented devm_pm_domain_attach() has 2 state >>>>>> arguments, one for the domain state on attach, one for the domain state on >>>>>> detach. >>>>> >>>>> dev_pm_domain_attach() is not part driver API and I'm not convinced at >>>> >>>> Is the concern that devm_pm_domain_attach() will be [ab]used by drivers? >>> >>> Yes, among other things. >> >> Maybe naming could make abuse at least obvious to spot? e.g. >> pm_domain_attach_with_devm_release() > > If I'm not mistaken, it is not even necessary to use devres for this. > > You might as well add a dev_pm_domain_detach() call to > device_unbind_cleanup() after devres_release_all(). There is a slight > complication related to the second argument of it, but I suppose that > this can be determined at the attach time and stored in a new device > PM flag, or similar. > I looked into this solution. I've tested it for all my failure cases and went good. > Note that dev->pm_domain is expected to be cleared by ->detach(), so > this should not cause the domain to be detached twice in a row from > the same device, but that needs to be double-checked. The genpd_dev_pm_detach() calls genpd_remove_device() -> dev_pm_domain_set(dev, NULL) which sets the dev->pm_domain = NULL. I can't find any other detach function in the current code base. The code I've tested for this solution is this one: diff --git a/drivers/base/dd.c b/drivers/base/dd.c index b526e0e0f52d..5e9750d007b4 100644 --- a/drivers/base/dd.c +++ b/drivers/base/dd.c @@ -25,6 +25,7 @@ #include <linux/kthread.h> #include <linux/wait.h> #include <linux/async.h> +#include <linux/pm_domain.h> #include <linux/pm_runtime.h> #include <linux/pinctrl/devinfo.h> #include <linux/slab.h> @@ -552,8 +553,11 @@ static void device_unbind_cleanup(struct device *dev) dev->dma_range_map = NULL; device_set_driver(dev, NULL); dev_set_drvdata(dev, NULL); - if (dev->pm_domain && dev->pm_domain->dismiss) - dev->pm_domain->dismiss(dev); + if (dev->pm_domain) { + if (dev->pm_domain->dismiss) + dev->pm_domain->dismiss(dev); + dev_pm_domain_detach(dev, dev->pm_domain->detach_power_off); + } pm_runtime_reinit(dev); dev_pm_set_driver_flags(dev, 0); } diff --git a/drivers/base/platform.c b/drivers/base/platform.c index 075ec1d1b73a..2459be6aecf4 100644 --- a/drivers/base/platform.c +++ b/drivers/base/platform.c @@ -1400,11 +1400,8 @@ static int platform_probe(struct device *_dev) if (ret) goto out; - if (drv->probe) { + if (drv->probe) ret = drv->probe(dev); - if (ret) - dev_pm_domain_detach(_dev, true); - } out: if (drv->prevent_deferred_probe && ret == -EPROBE_DEFER) { @@ -1422,7 +1419,6 @@ static void platform_remove(struct device *_dev) if (drv->remove) drv->remove(dev); - dev_pm_domain_detach(_dev, true); } static void platform_shutdown(struct device *_dev) diff --git a/drivers/base/power/common.c b/drivers/base/power/common.c index 781968a128ff..4bd1e3c7f401 100644 --- a/drivers/base/power/common.c +++ b/drivers/base/power/common.c @@ -111,6 +111,9 @@ int dev_pm_domain_attach(struct device *dev, bool power_on) if (!ret) ret = genpd_dev_pm_attach(dev); + if (dev->pm_domain) + dev->pm_domain->detach_power_off = power_on; + return ret < 0 ? ret : 0; } EXPORT_SYMBOL_GPL(dev_pm_domain_attach); diff --git a/include/linux/pm.h b/include/linux/pm.h index f0bd8fbae4f2..12e97e09e85c 100644 --- a/include/linux/pm.h +++ b/include/linux/pm.h @@ -748,6 +748,7 @@ struct dev_pm_domain { void (*sync)(struct device *dev); void (*dismiss)(struct device *dev); int (*set_performance_state)(struct device *dev, unsigned int state); + bool detach_power_off; }; Rafael, Ulf, Dmitry, Jonathan, all, Could you please let me know how do you consider this approach? Thank you, Claudiu
On Fri, Jun 13, 2025 at 9:39 AM Claudiu Beznea <claudiu.beznea@tuxon.dev> wrote: > > Hi, Rafael, > > On 09.06.2025 22:59, Rafael J. Wysocki wrote: > > On Sat, Jun 7, 2025 at 3:06 PM Jonathan Cameron <jic23@kernel.org> wrote: > >> > >> On Fri, 6 Jun 2025 22:01:52 +0200 > >> "Rafael J. Wysocki" <rafael@kernel.org> wrote: > >> > >> Hi Rafael, > >> > >>> On Fri, Jun 6, 2025 at 8:55 PM Dmitry Torokhov > >>> <dmitry.torokhov@gmail.com> wrote: > >>>> > >>>> On Fri, Jun 06, 2025 at 06:00:34PM +0200, Rafael J. Wysocki wrote: > >>>>> On Fri, Jun 6, 2025 at 1:18 PM Claudiu <claudiu.beznea@tuxon.dev> wrote: > >>>>>> > >>>>>> From: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com> > >>>>>> > >>>>>> The dev_pm_domain_attach() function is typically used in bus code alongside > >>>>>> dev_pm_domain_detach(), often following patterns like: > >>>>>> > >>>>>> static int bus_probe(struct device *_dev) > >>>>>> { > >>>>>> struct bus_driver *drv = to_bus_driver(dev->driver); > >>>>>> struct bus_device *dev = to_bus_device(_dev); > >>>>>> int ret; > >>>>>> > >>>>>> // ... > >>>>>> > >>>>>> ret = dev_pm_domain_attach(_dev, true); > >>>>>> if (ret) > >>>>>> return ret; > >>>>>> > >>>>>> if (drv->probe) > >>>>>> ret = drv->probe(dev); > >>>>>> > >>>>>> // ... > >>>>>> } > >>>>>> > >>>>>> static void bus_remove(struct device *_dev) > >>>>>> { > >>>>>> struct bus_driver *drv = to_bus_driver(dev->driver); > >>>>>> struct bus_device *dev = to_bus_device(_dev); > >>>>>> > >>>>>> if (drv->remove) > >>>>>> drv->remove(dev); > >>>>>> dev_pm_domain_detach(_dev); > >>>>>> } > >>>>>> > >>>>>> When the driver's probe function uses devres-managed resources that depend > >>>>>> on the power domain state, those resources are released later during > >>>>>> device_unbind_cleanup(). > >>>>>> > >>>>>> Releasing devres-managed resources that depend on the power domain state > >>>>>> after detaching the device from its PM domain can cause failures. > >>>>>> > >>>>>> For example, if the driver uses devm_pm_runtime_enable() in its probe > >>>>>> function, and the device's clocks are managed by the PM domain, then > >>>>>> during removal the runtime PM is disabled in device_unbind_cleanup() after > >>>>>> the clocks have been removed from the PM domain. It may happen that the > >>>>>> devm_pm_runtime_enable() action causes the device to be runtime-resumed. > >>>>> > >>>>> Don't use devm_pm_runtime_enable() then. > >>>> > >>>> What about other devm_ APIs? Are you suggesting that platform drivers > >>>> should not be using devm_clk*(), devm_regulator_*(), > >>>> devm_request_*_irq() and devm_add_action_or_reset()? Because again, > >>>> dev_pm_domain_detach() that is called by platform bus_remove() may shut > >>>> off the device too early, before cleanup code has a chance to execute > >>>> proper cleanup. > >>>> > >>>> The issue is not limited to runtime PM. > >>>> > >>>>> > >>>>>> If the driver specific runtime PM APIs access registers directly, this > >>>>>> will lead to accessing device registers without clocks being enabled. > >>>>>> Similar issues may occur with other devres actions that access device > >>>>>> registers. > >>>>>> > >>>>>> Add devm_pm_domain_attach(). When replacing the dev_pm_domain_attach() and > >>>>>> dev_pm_domain_detach() in bus probe and bus remove, it ensures that the > >>>>>> device is detached from its PM domain in device_unbind_cleanup(), only > >>>>>> after all driver's devres-managed resources have been release. > >>>>>> > >>>>>> For flexibility, the implemented devm_pm_domain_attach() has 2 state > >>>>>> arguments, one for the domain state on attach, one for the domain state on > >>>>>> detach. > >>>>> > >>>>> dev_pm_domain_attach() is not part driver API and I'm not convinced at > >>>> > >>>> Is the concern that devm_pm_domain_attach() will be [ab]used by drivers? > >>> > >>> Yes, among other things. > >> > >> Maybe naming could make abuse at least obvious to spot? e.g. > >> pm_domain_attach_with_devm_release() > > > > If I'm not mistaken, it is not even necessary to use devres for this. > > > > You might as well add a dev_pm_domain_detach() call to > > device_unbind_cleanup() after devres_release_all(). There is a slight > > complication related to the second argument of it, but I suppose that > > this can be determined at the attach time and stored in a new device > > PM flag, or similar. > > > > I looked into this solution. I've tested it for all my failure cases and > went good. OK > > Note that dev->pm_domain is expected to be cleared by ->detach(), so > > this should not cause the domain to be detached twice in a row from > > the same device, but that needs to be double-checked. > > The genpd_dev_pm_detach() calls genpd_remove_device() -> > dev_pm_domain_set(dev, NULL) which sets the dev->pm_domain = NULL. I can't > find any other detach function in the current code base. There is also acpi_dev_pm_detach() which can be somewhat hard to find, but it calls dev_pm_domain_set(dev, NULL) either. > The code I've tested for this solution is this one: > > diff --git a/drivers/base/dd.c b/drivers/base/dd.c > index b526e0e0f52d..5e9750d007b4 100644 > --- a/drivers/base/dd.c > +++ b/drivers/base/dd.c > @@ -25,6 +25,7 @@ > #include <linux/kthread.h> > #include <linux/wait.h> > #include <linux/async.h> > +#include <linux/pm_domain.h> > #include <linux/pm_runtime.h> > #include <linux/pinctrl/devinfo.h> > #include <linux/slab.h> > @@ -552,8 +553,11 @@ static void device_unbind_cleanup(struct device *dev) > dev->dma_range_map = NULL; > device_set_driver(dev, NULL); > dev_set_drvdata(dev, NULL); > - if (dev->pm_domain && dev->pm_domain->dismiss) > - dev->pm_domain->dismiss(dev); > + if (dev->pm_domain) { > + if (dev->pm_domain->dismiss) > + dev->pm_domain->dismiss(dev); > + dev_pm_domain_detach(dev, dev->pm_domain->detach_power_off); I would do the "detach" before the "dismiss" to retain the current ordering. Also it is interesting that you ended up calling them both in one place. It kind of indicates that the PM domains attached via dev_pm_domain_attach() should be attached earlier and just use the ->activate() and ->dismiss() callbacks. > + } > pm_runtime_reinit(dev); > dev_pm_set_driver_flags(dev, 0); > } > diff --git a/drivers/base/platform.c b/drivers/base/platform.c > index 075ec1d1b73a..2459be6aecf4 100644 > --- a/drivers/base/platform.c > +++ b/drivers/base/platform.c > @@ -1400,11 +1400,8 @@ static int platform_probe(struct device *_dev) > if (ret) > goto out; > > - if (drv->probe) { > + if (drv->probe) > ret = drv->probe(dev); > - if (ret) > - dev_pm_domain_detach(_dev, true); > - } > > out: > if (drv->prevent_deferred_probe && ret == -EPROBE_DEFER) { > @@ -1422,7 +1419,6 @@ static void platform_remove(struct device *_dev) > > if (drv->remove) > drv->remove(dev); > - dev_pm_domain_detach(_dev, true); > } > > static void platform_shutdown(struct device *_dev) > diff --git a/drivers/base/power/common.c b/drivers/base/power/common.c > index 781968a128ff..4bd1e3c7f401 100644 > --- a/drivers/base/power/common.c > +++ b/drivers/base/power/common.c > @@ -111,6 +111,9 @@ int dev_pm_domain_attach(struct device *dev, bool power_on) > if (!ret) > ret = genpd_dev_pm_attach(dev); > > + if (dev->pm_domain) > + dev->pm_domain->detach_power_off = power_on; This will not work for acpi_general_pm_domain because it is shared by all of its users. It is likely to not work for shared PM domains in general. I would put the new flag into struct dev_pm_info. > + > return ret < 0 ? ret : 0; > } > EXPORT_SYMBOL_GPL(dev_pm_domain_attach); > diff --git a/include/linux/pm.h b/include/linux/pm.h > index f0bd8fbae4f2..12e97e09e85c 100644 > --- a/include/linux/pm.h > +++ b/include/linux/pm.h > @@ -748,6 +748,7 @@ struct dev_pm_domain { > void (*sync)(struct device *dev); > void (*dismiss)(struct device *dev); > int (*set_performance_state)(struct device *dev, unsigned int state); > + bool detach_power_off; > }; > > Rafael, Ulf, Dmitry, Jonathan, all, > > Could you please let me know how do you consider this approach? Please see my comments above. Thanks!
Hi, Rafael, On 13.06.2025 13:02, Rafael J. Wysocki wrote: > On Fri, Jun 13, 2025 at 9:39 AM Claudiu Beznea <claudiu.beznea@tuxon.dev> wrote: >> >> Hi, Rafael, >> >> On 09.06.2025 22:59, Rafael J. Wysocki wrote: >>> On Sat, Jun 7, 2025 at 3:06 PM Jonathan Cameron <jic23@kernel.org> wrote: >>>> >>>> On Fri, 6 Jun 2025 22:01:52 +0200 >>>> "Rafael J. Wysocki" <rafael@kernel.org> wrote: >>>> >>>> Hi Rafael, >>>> >>>>> On Fri, Jun 6, 2025 at 8:55 PM Dmitry Torokhov >>>>> <dmitry.torokhov@gmail.com> wrote: >>>>>> >>>>>> On Fri, Jun 06, 2025 at 06:00:34PM +0200, Rafael J. Wysocki wrote: >>>>>>> On Fri, Jun 6, 2025 at 1:18 PM Claudiu <claudiu.beznea@tuxon.dev> wrote: >>>>>>>> >>>>>>>> From: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com> >>>>>>>> >>>>>>>> The dev_pm_domain_attach() function is typically used in bus code alongside >>>>>>>> dev_pm_domain_detach(), often following patterns like: >>>>>>>> >>>>>>>> static int bus_probe(struct device *_dev) >>>>>>>> { >>>>>>>> struct bus_driver *drv = to_bus_driver(dev->driver); >>>>>>>> struct bus_device *dev = to_bus_device(_dev); >>>>>>>> int ret; >>>>>>>> >>>>>>>> // ... >>>>>>>> >>>>>>>> ret = dev_pm_domain_attach(_dev, true); >>>>>>>> if (ret) >>>>>>>> return ret; >>>>>>>> >>>>>>>> if (drv->probe) >>>>>>>> ret = drv->probe(dev); >>>>>>>> >>>>>>>> // ... >>>>>>>> } >>>>>>>> >>>>>>>> static void bus_remove(struct device *_dev) >>>>>>>> { >>>>>>>> struct bus_driver *drv = to_bus_driver(dev->driver); >>>>>>>> struct bus_device *dev = to_bus_device(_dev); >>>>>>>> >>>>>>>> if (drv->remove) >>>>>>>> drv->remove(dev); >>>>>>>> dev_pm_domain_detach(_dev); >>>>>>>> } >>>>>>>> >>>>>>>> When the driver's probe function uses devres-managed resources that depend >>>>>>>> on the power domain state, those resources are released later during >>>>>>>> device_unbind_cleanup(). >>>>>>>> >>>>>>>> Releasing devres-managed resources that depend on the power domain state >>>>>>>> after detaching the device from its PM domain can cause failures. >>>>>>>> >>>>>>>> For example, if the driver uses devm_pm_runtime_enable() in its probe >>>>>>>> function, and the device's clocks are managed by the PM domain, then >>>>>>>> during removal the runtime PM is disabled in device_unbind_cleanup() after >>>>>>>> the clocks have been removed from the PM domain. It may happen that the >>>>>>>> devm_pm_runtime_enable() action causes the device to be runtime-resumed. >>>>>>> >>>>>>> Don't use devm_pm_runtime_enable() then. >>>>>> >>>>>> What about other devm_ APIs? Are you suggesting that platform drivers >>>>>> should not be using devm_clk*(), devm_regulator_*(), >>>>>> devm_request_*_irq() and devm_add_action_or_reset()? Because again, >>>>>> dev_pm_domain_detach() that is called by platform bus_remove() may shut >>>>>> off the device too early, before cleanup code has a chance to execute >>>>>> proper cleanup. >>>>>> >>>>>> The issue is not limited to runtime PM. >>>>>> >>>>>>> >>>>>>>> If the driver specific runtime PM APIs access registers directly, this >>>>>>>> will lead to accessing device registers without clocks being enabled. >>>>>>>> Similar issues may occur with other devres actions that access device >>>>>>>> registers. >>>>>>>> >>>>>>>> Add devm_pm_domain_attach(). When replacing the dev_pm_domain_attach() and >>>>>>>> dev_pm_domain_detach() in bus probe and bus remove, it ensures that the >>>>>>>> device is detached from its PM domain in device_unbind_cleanup(), only >>>>>>>> after all driver's devres-managed resources have been release. >>>>>>>> >>>>>>>> For flexibility, the implemented devm_pm_domain_attach() has 2 state >>>>>>>> arguments, one for the domain state on attach, one for the domain state on >>>>>>>> detach. >>>>>>> >>>>>>> dev_pm_domain_attach() is not part driver API and I'm not convinced at >>>>>> >>>>>> Is the concern that devm_pm_domain_attach() will be [ab]used by drivers? >>>>> >>>>> Yes, among other things. >>>> >>>> Maybe naming could make abuse at least obvious to spot? e.g. >>>> pm_domain_attach_with_devm_release() >>> >>> If I'm not mistaken, it is not even necessary to use devres for this. >>> >>> You might as well add a dev_pm_domain_detach() call to >>> device_unbind_cleanup() after devres_release_all(). There is a slight >>> complication related to the second argument of it, but I suppose that >>> this can be determined at the attach time and stored in a new device >>> PM flag, or similar. >>> >> >> I looked into this solution. I've tested it for all my failure cases and >> went good. > > OK > >>> Note that dev->pm_domain is expected to be cleared by ->detach(), so >>> this should not cause the domain to be detached twice in a row from >>> the same device, but that needs to be double-checked. >> >> The genpd_dev_pm_detach() calls genpd_remove_device() -> >> dev_pm_domain_set(dev, NULL) which sets the dev->pm_domain = NULL. I can't >> find any other detach function in the current code base. > > There is also acpi_dev_pm_detach() which can be somewhat hard to find, > but it calls dev_pm_domain_set(dev, NULL) either. Thank you for the pointer. > >> The code I've tested for this solution is this one: >> >> diff --git a/drivers/base/dd.c b/drivers/base/dd.c >> index b526e0e0f52d..5e9750d007b4 100644 >> --- a/drivers/base/dd.c >> +++ b/drivers/base/dd.c >> @@ -25,6 +25,7 @@ >> #include <linux/kthread.h> >> #include <linux/wait.h> >> #include <linux/async.h> >> +#include <linux/pm_domain.h> >> #include <linux/pm_runtime.h> >> #include <linux/pinctrl/devinfo.h> >> #include <linux/slab.h> >> @@ -552,8 +553,11 @@ static void device_unbind_cleanup(struct device *dev) >> dev->dma_range_map = NULL; >> device_set_driver(dev, NULL); >> dev_set_drvdata(dev, NULL); >> - if (dev->pm_domain && dev->pm_domain->dismiss) >> - dev->pm_domain->dismiss(dev); >> + if (dev->pm_domain) { >> + if (dev->pm_domain->dismiss) >> + dev->pm_domain->dismiss(dev); >> + dev_pm_domain_detach(dev, dev->pm_domain->detach_power_off); > > I would do the "detach" before the "dismiss" to retain the current ordering. OK. > > Also it is interesting that you ended up calling them both in one > place. It kind of indicates that the PM domains attached via > dev_pm_domain_attach() should be attached earlier and just use the > ->activate() and ->dismiss() callbacks. > >> + } >> pm_runtime_reinit(dev); >> dev_pm_set_driver_flags(dev, 0); >> } >> diff --git a/drivers/base/platform.c b/drivers/base/platform.c >> index 075ec1d1b73a..2459be6aecf4 100644 >> --- a/drivers/base/platform.c >> +++ b/drivers/base/platform.c >> @@ -1400,11 +1400,8 @@ static int platform_probe(struct device *_dev) >> if (ret) >> goto out; >> >> - if (drv->probe) { >> + if (drv->probe) >> ret = drv->probe(dev); >> - if (ret) >> - dev_pm_domain_detach(_dev, true); >> - } >> >> out: >> if (drv->prevent_deferred_probe && ret == -EPROBE_DEFER) { >> @@ -1422,7 +1419,6 @@ static void platform_remove(struct device *_dev) >> >> if (drv->remove) >> drv->remove(dev); >> - dev_pm_domain_detach(_dev, true); >> } >> >> static void platform_shutdown(struct device *_dev) >> diff --git a/drivers/base/power/common.c b/drivers/base/power/common.c >> index 781968a128ff..4bd1e3c7f401 100644 >> --- a/drivers/base/power/common.c >> +++ b/drivers/base/power/common.c >> @@ -111,6 +111,9 @@ int dev_pm_domain_attach(struct device *dev, bool power_on) >> if (!ret) >> ret = genpd_dev_pm_attach(dev); >> >> + if (dev->pm_domain) >> + dev->pm_domain->detach_power_off = power_on; > > This will not work for acpi_general_pm_domain because it is shared by > all of its users. > > It is likely to not work for shared PM domains in general. > > I would put the new flag into struct dev_pm_info. OK, I'll do it. Thank you for your input, Claudiu > >> + >> return ret < 0 ? ret : 0; >> } >> EXPORT_SYMBOL_GPL(dev_pm_domain_attach); >> diff --git a/include/linux/pm.h b/include/linux/pm.h >> index f0bd8fbae4f2..12e97e09e85c 100644 >> --- a/include/linux/pm.h >> +++ b/include/linux/pm.h >> @@ -748,6 +748,7 @@ struct dev_pm_domain { >> void (*sync)(struct device *dev); >> void (*dismiss)(struct device *dev); >> int (*set_performance_state)(struct device *dev, unsigned int state); >> + bool detach_power_off; >> }; >> >> Rafael, Ulf, Dmitry, Jonathan, all, >> >> Could you please let me know how do you consider this approach? > > Please see my comments above. > > Thanks!
From: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com> Hi, As a result of discussion at [1], series adds the devm_pm_domain_attach() and uses it in platform bus probe. Please provide your feedback. Thank you, Claudiu [1] https://lore.kernel.org/all/20250215130849.227812-1-claudiu.beznea.uj@bp.renesas.com Changes in v3: - dropped the detach_power_off argument of devm_pm_domain_attach() - use a single cleanup function - fixed build warning Changes in v2: - add devm_pm_domain_attach() - drop the devres grup open/close approach and use the newly added devm_pm_domain_attach() Claudiu Beznea (2): PM: domains: Add devres variant for dev_pm_domain_attach() driver core: platform: Use devm_pm_domain_attach() drivers/base/platform.c | 8 ++---- drivers/base/power/common.c | 50 +++++++++++++++++++++++++++++++++++++ include/linux/pm_domain.h | 6 +++++ 3 files changed, 58 insertions(+), 6 deletions(-)