Hi Greg & all,
Commit d12544fb2aa9 ("PM: runtime: Remove link state checks in
rpm_get/put_supplier()") merged recently introduced a weakness
in the handling of device links in the runtime PM framework that
may be confusing and even harmful.
Namely, the checks removed by that commit prevented PM-runtime from
getting or dropping references to the supplier device whose driver
was going away via its links to consumers, which specifically allowed
the pm_runtime_clean_up_links() called from __device_release_driver()
to run without interfering with runtime suspend/resume of consumer
devices (which still might happen even though the drivers had been
unbound from them by that time).
After the above commit, calling pm_runtime_clean_up_links() from
__device_release_driver() makes a little sense and it may be interfering
destructively with regular PM-runtime suspend/resume control flows, so
it needs to be either fixed or dropped altogether. I prefer the latter,
because among other things this removes an arbitrary difference in the
handling of managed device links with respect to the stateless ones,
so patch [2/3] is doing just that.
However, in some rare cases pm_runtime_clean_up_links() may help to clean
up leftover PM-runtime references, so if that function goes away, they
need to be cleaned up elsewhere. That's why patch [1/3] modifies
__device_link_del() to drop them upon device link removal (which also
needs to be done for stateless device links and that's why I'm regarding
this patch as a fix).
Finally, to avoid pointless overhead related to suspending and resuming
the target device for multiple times in a row in __device_release_driver(),
it is better to resume it upfront before checking its links to consumers,
which is done by patch [3/3].
While this series touches the driver core, it really is mostly related to
runtime PM, so I can apply it if that's OK.
Thanks!
From: Rafael J. Wysocki <[email protected]>
After commit d12544fb2aa9 ("PM: runtime: Remove link state checks in
rpm_get/put_supplier()") nothing prevents the consumer device's
runtime PM from acquiring additional references to the supplier
device after pm_runtime_clean_up_links() has run (or even while it
is running), so calling this function from __device_release_driver()
may be pointless (or even harmful).
Moreover, it ignores stateless device links, so the runtime PM
handling of managed and stateless device links is inconsistent
because of it, so better get rid of it entirely.
Fixes: d12544fb2aa9 ("PM: runtime: Remove link state checks in rpm_get/put_supplier()")
Signed-off-by: Rafael J. Wysocki <[email protected]>
Cc: 5.1+ <[email protected]> # 5.1+
---
drivers/base/dd.c | 1 -
drivers/base/power/runtime.c | 36 ------------------------------------
include/linux/pm_runtime.h | 2 --
3 files changed, 39 deletions(-)
Index: linux-pm/drivers/base/dd.c
===================================================================
--- linux-pm.orig/drivers/base/dd.c
+++ linux-pm/drivers/base/dd.c
@@ -1133,7 +1133,6 @@ static void __device_release_driver(stru
}
pm_runtime_get_sync(dev);
- pm_runtime_clean_up_links(dev);
driver_sysfs_remove(dev);
Index: linux-pm/drivers/base/power/runtime.c
===================================================================
--- linux-pm.orig/drivers/base/power/runtime.c
+++ linux-pm/drivers/base/power/runtime.c
@@ -1643,42 +1643,6 @@ void pm_runtime_remove(struct device *de
}
/**
- * pm_runtime_clean_up_links - Prepare links to consumers for driver removal.
- * @dev: Device whose driver is going to be removed.
- *
- * Check links from this device to any consumers and if any of them have active
- * runtime PM references to the device, drop the usage counter of the device
- * (as many times as needed).
- *
- * Links with the DL_FLAG_MANAGED flag unset are ignored.
- *
- * Since the device is guaranteed to be runtime-active at the point this is
- * called, nothing else needs to be done here.
- *
- * Moreover, this is called after device_links_busy() has returned 'false', so
- * the status of each link is guaranteed to be DL_STATE_SUPPLIER_UNBIND and
- * therefore rpm_active can't be manipulated concurrently.
- */
-void pm_runtime_clean_up_links(struct device *dev)
-{
- struct device_link *link;
- int idx;
-
- idx = device_links_read_lock();
-
- list_for_each_entry_rcu(link, &dev->links.consumers, s_node,
- device_links_read_lock_held()) {
- if (!(link->flags & DL_FLAG_MANAGED))
- continue;
-
- while (refcount_dec_not_one(&link->rpm_active))
- pm_runtime_put_noidle(dev);
- }
-
- device_links_read_unlock(idx);
-}
-
-/**
* pm_runtime_get_suppliers - Resume and reference-count supplier devices.
* @dev: Consumer device.
*/
Index: linux-pm/include/linux/pm_runtime.h
===================================================================
--- linux-pm.orig/include/linux/pm_runtime.h
+++ linux-pm/include/linux/pm_runtime.h
@@ -54,7 +54,6 @@ extern u64 pm_runtime_autosuspend_expira
extern void pm_runtime_update_max_time_suspended(struct device *dev,
s64 delta_ns);
extern void pm_runtime_set_memalloc_noio(struct device *dev, bool enable);
-extern void pm_runtime_clean_up_links(struct device *dev);
extern void pm_runtime_get_suppliers(struct device *dev);
extern void pm_runtime_put_suppliers(struct device *dev);
extern void pm_runtime_new_link(struct device *dev);
@@ -276,7 +275,6 @@ static inline u64 pm_runtime_autosuspend
struct device *dev) { return 0; }
static inline void pm_runtime_set_memalloc_noio(struct device *dev,
bool enable){}
-static inline void pm_runtime_clean_up_links(struct device *dev) {}
static inline void pm_runtime_get_suppliers(struct device *dev) {}
static inline void pm_runtime_put_suppliers(struct device *dev) {}
static inline void pm_runtime_new_link(struct device *dev) {}
On Friday, October 23, 2020 5:50:04 AM CEST chenxiang (M) wrote:
> Hi Rafael,
>
> 在 2020/10/22 3:10, Rafael J. Wysocki 写道:
> > Hi Greg & all,
> >
> > Commit d12544fb2aa9 ("PM: runtime: Remove link state checks in
> > rpm_get/put_supplier()") merged recently introduced a weakness
> > in the handling of device links in the runtime PM framework that
> > may be confusing and even harmful.
> >
> > Namely, the checks removed by that commit prevented PM-runtime from
> > getting or dropping references to the supplier device whose driver
> > was going away via its links to consumers, which specifically allowed
> > the pm_runtime_clean_up_links() called from __device_release_driver()
> > to run without interfering with runtime suspend/resume of consumer
> > devices (which still might happen even though the drivers had been
> > unbound from them by that time).
> >
> > After the above commit, calling pm_runtime_clean_up_links() from
> > __device_release_driver() makes a little sense and it may be interfering
> > destructively with regular PM-runtime suspend/resume control flows, so
> > it needs to be either fixed or dropped altogether. I prefer the latter,
> > because among other things this removes an arbitrary difference in the
> > handling of managed device links with respect to the stateless ones,
> > so patch [2/3] is doing just that.
> >
> > However, in some rare cases pm_runtime_clean_up_links() may help to clean
> > up leftover PM-runtime references, so if that function goes away, they
> > need to be cleaned up elsewhere. That's why patch [1/3] modifies
> > __device_link_del() to drop them upon device link removal (which also
> > needs to be done for stateless device links and that's why I'm regarding
> > this patch as a fix).
> >
> > Finally, to avoid pointless overhead related to suspending and resuming
> > the target device for multiple times in a row in __device_release_driver(),
> > it is better to resume it upfront before checking its links to consumers,
> > which is done by patch [3/3].
>
>
> I have tested the patchset, and it solves my reported issue, so please
> feel free to add :
> Tested-by: Xiang Chen <[email protected]>
Thank you!
Hi Greg,
On Wed, Oct 21, 2020 at 9:14 PM Rafael J. Wysocki <[email protected]> wrote:
>
> Hi Greg & all,
>
> Commit d12544fb2aa9 ("PM: runtime: Remove link state checks in
> rpm_get/put_supplier()") merged recently introduced a weakness
> in the handling of device links in the runtime PM framework that
> may be confusing and even harmful.
>
> Namely, the checks removed by that commit prevented PM-runtime from
> getting or dropping references to the supplier device whose driver
> was going away via its links to consumers, which specifically allowed
> the pm_runtime_clean_up_links() called from __device_release_driver()
> to run without interfering with runtime suspend/resume of consumer
> devices (which still might happen even though the drivers had been
> unbound from them by that time).
>
> After the above commit, calling pm_runtime_clean_up_links() from
> __device_release_driver() makes a little sense and it may be interfering
> destructively with regular PM-runtime suspend/resume control flows, so
> it needs to be either fixed or dropped altogether. I prefer the latter,
> because among other things this removes an arbitrary difference in the
> handling of managed device links with respect to the stateless ones,
> so patch [2/3] is doing just that.
>
> However, in some rare cases pm_runtime_clean_up_links() may help to clean
> up leftover PM-runtime references, so if that function goes away, they
> need to be cleaned up elsewhere. That's why patch [1/3] modifies
> __device_link_del() to drop them upon device link removal (which also
> needs to be done for stateless device links and that's why I'm regarding
> this patch as a fix).
>
> Finally, to avoid pointless overhead related to suspending and resuming
> the target device for multiple times in a row in __device_release_driver(),
> it is better to resume it upfront before checking its links to consumers,
> which is done by patch [3/3].
>
> While this series touches the driver core, it really is mostly related to
> runtime PM, so I can apply it if that's OK.
Any concerns regarding this series?
If not, I'd like to queue it up for -rc3, because the current behavior
in there is quite confusing (or worse).
Cheers!
On Wed, Oct 21, 2020 at 09:10:08PM +0200, Rafael J. Wysocki wrote:
> Hi Greg & all,
>
> Commit d12544fb2aa9 ("PM: runtime: Remove link state checks in
> rpm_get/put_supplier()") merged recently introduced a weakness
> in the handling of device links in the runtime PM framework that
> may be confusing and even harmful.
>
> Namely, the checks removed by that commit prevented PM-runtime from
> getting or dropping references to the supplier device whose driver
> was going away via its links to consumers, which specifically allowed
> the pm_runtime_clean_up_links() called from __device_release_driver()
> to run without interfering with runtime suspend/resume of consumer
> devices (which still might happen even though the drivers had been
> unbound from them by that time).
>
> After the above commit, calling pm_runtime_clean_up_links() from
> __device_release_driver() makes a little sense and it may be interfering
> destructively with regular PM-runtime suspend/resume control flows, so
> it needs to be either fixed or dropped altogether. I prefer the latter,
> because among other things this removes an arbitrary difference in the
> handling of managed device links with respect to the stateless ones,
> so patch [2/3] is doing just that.
>
> However, in some rare cases pm_runtime_clean_up_links() may help to clean
> up leftover PM-runtime references, so if that function goes away, they
> need to be cleaned up elsewhere. That's why patch [1/3] modifies
> __device_link_del() to drop them upon device link removal (which also
> needs to be done for stateless device links and that's why I'm regarding
> this patch as a fix).
>
> Finally, to avoid pointless overhead related to suspending and resuming
> the target device for multiple times in a row in __device_release_driver(),
> it is better to resume it upfront before checking its links to consumers,
> which is done by patch [3/3].
>
> While this series touches the driver core, it really is mostly related to
> runtime PM, so I can apply it if that's OK.
Please do, sorry for the delay in reviewing them:
Reviewed-by: Greg Kroah-Hartman <[email protected]>