PM runtime auto-cleanup macros
Subsystems (2)
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the __free(pm_runtime_put) in commit bfa4477751e9, and there is a (single) user of it in pci-sysfs.c.
Now I wanted to extend it to pm_runtime_put_autosuspend() as:
DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))
Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;
that is similar as done in pci-sysfs.c. So far, so good.
But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
#define pm_runtime_auto_clean(dev, var) \
struct device *var __free(pm_runtime_put) = (dev)
#define pm_runtime_auto_clean_autosuspend(dev, var) \
struct device *var __free(pm_runtime_put_autosuspend) = (dev)
and the code will be like:
pm_runtime_get_sync(dev); pm_runtime_auto_clean(dev, pmdev);
or
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; pm_runtime_auto_clean_autosuspend(dev, pmdev);
Alternatively, we may define a class, e.g.
CLASS(pm_runtime_resume_and_get, pmdev); if (pmdev.ret < 0) return pmdev.ret;
but it'll be a bit more code to define the full class, and the get*() and put*() combination would be fixed with this approach -- which is a downside.
All above are an idea for now. Let me know if I should go further along with this, or there is already a better another approach.
(And the macros can be better named, sure :)
thanks,
Takashi
Quoted 21 lines · Takashi Iwai
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the
__free(pm_runtime_put)in commitbfa4477751e9, and there is a (single) user of it inpci-sysfs.c.Now I wanted to extend it to
pm_runtime_put_autosuspend()as:DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;
Quoted 8 lines
that is similar as done in
pci-sysfs.c. So far, so good.But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
#define
pm_runtime_auto_clean(dev, var)\
struct device *var__free(pm_runtime_put)= (dev)
It can be even simpler by assigning a temporary variable such as:
#define pm_runtime_auto_clean(dev) \ struct device *__pm_runtime_var ## __LINE__ __free(pm_runtime_put) = (dev)
Takashi
Quoted 34 lines
#define
pm_runtime_auto_clean_autosuspend(dev, var)\
struct device *var__free(pm_runtime_put_autosuspend)= (dev)and the code will be like:
pm_runtime_get_sync(dev); pm_runtime_auto_clean(dev, pmdev);or
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; pm_runtime_auto_clean_autosuspend(dev, pmdev);Alternatively, we may define a class, e.g.
CLASS(pm_runtime_resume_and_get, pmdev); if (pmdev.ret < 0) return pmdev.ret;but it'll be a bit more code to define the full class, and the get*() and put*() combination would be fixed with this approach -- which is a downside.
All above are an idea for now. Let me know if I should go further along with this, or there is already a better another approach.
(And the macros can be better named, sure :)
thanks,
Takashi
Rafael J. Wysocki
maintainerHi,
Sorry for the delay.
Quoted 30 lines · Takashi Iwai
Quoted 27 lines · Takashi Iwai
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the
__free(pm_runtime_put)in commitbfa4477751e9, and there is a (single) user of it inpci-sysfs.c.Now I wanted to extend it to
pm_runtime_put_autosuspend()as:DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;that is similar as done in
pci-sysfs.c. So far, so good.But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
Quoted 7 lines
Quoted 2 lines
#define
pm_runtime_auto_clean(dev, var)\
struct device *var__free(pm_runtime_put)= (dev)It can be even simpler by assigning a temporary variable such as:
#define pm_runtime_auto_clean(dev) \ struct device *__pm_runtime_var ## __LINE__ __free(pm_runtime_put) = (dev)
Well, if there's something like
struct device *pm_runtime_resume_and_get_dev(struct device *dev)
{
int ret = pm_runtime_resume_and_get(dev);
if (ret < 0)
return ERR_PTR(ret);
return dev;
}It would be a matter of redefining the FREE to also take error pointers into account and you could do
struct device *__dev __free(pm_runtim_put) = pm_runtime_resume_and_get_dev(dev); if (IS_ERR(__dev))
return PTR_ERR(__dev);
Quoted 55 lines · Rafael J. Wysocki
Hi,
Sorry for the delay.
Quoted 30 lines · Takashi Iwai
Quoted 27 lines · Takashi Iwai
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the
__free(pm_runtime_put)in commitbfa4477751e9, and there is a (single) user of it inpci-sysfs.c.Now I wanted to extend it to
pm_runtime_put_autosuspend()as:DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;that is similar as done in
pci-sysfs.c. So far, so good.But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
Quoted 7 lines
Quoted 2 lines
#define
pm_runtime_auto_clean(dev, var)\
struct device *var__free(pm_runtime_put)= (dev)It can be even simpler by assigning a temporary variable such as:
#define pm_runtime_auto_clean(dev) \ struct device *__pm_runtime_var ## __LINE__ __free(pm_runtime_put) = (dev)Well, if there's something like
struct device *pm_runtime_resume_and_get_dev(struct device *dev)
{ int ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ERR_PTR(ret); return dev; }
Quoted 6 lines
It would be a matter of redefining the FREE to also take error pointers into account and you could do
struct device
*__dev__free(pm_runtim_put)=pm_runtime_resume_and_get_dev(dev); if (IS_ERR(__dev))return PTR_ERR(__dev);
That'll work, too. Though, I find the notion of __free() and a temporary variable __dev a bit too cumbersome; it's used only for auto-clean stuff, so it could be somewhat anonymous.
But it's all about a matter of taste, and I'd follow what you and other guys suggest.
FWIW, there are lots of code doing like
pm_runtime_get_sync(dev); mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put(dev); return;
or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put_autosuspend(dev); return 0;
and they can be converted nicely with guard() once when PM runtime can be automatically unreferenced. With my proposed change, it would become like:
pm_runtime_get_sync(dev); pm_runtime_auto_clean(dev); guard(mutex)(&foo); .... return;
or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; pm_runtime_auto_clean_autosuspend(dev); guard(mutex)(&foo); .... return 0;
thanks,
Takashi
Rafael J. Wysocki
maintainerQuoted 61 lines · Takashi Iwai
Quoted 58 lines · Rafael J. Wysocki
Hi,
Sorry for the delay.
Quoted 30 lines · Takashi Iwai
Quoted 27 lines · Takashi Iwai
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the
__free(pm_runtime_put)in commitbfa4477751e9, and there is a (single) user of it inpci-sysfs.c.Now I wanted to extend it to
pm_runtime_put_autosuspend()as:DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;that is similar as done in
pci-sysfs.c. So far, so good.But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
Quoted 7 lines
Quoted 2 lines
#define
pm_runtime_auto_clean(dev, var)\
struct device *var__free(pm_runtime_put)= (dev)It can be even simpler by assigning a temporary variable such as:
#define pm_runtime_auto_clean(dev) \ struct device *__pm_runtime_var ## __LINE__ __free(pm_runtime_put) = (dev)Well, if there's something like
struct device *pm_runtime_resume_and_get_dev(struct device *dev)
{ int ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ERR_PTR(ret); return dev; }It would be a matter of redefining the FREE to also take error pointers into account and you could do
Quoted 7 lines
Quoted 3 lines
struct device
*__dev__free(pm_runtim_put)=pm_runtime_resume_and_get_dev(dev); if (IS_ERR(__dev))return PTR_ERR(__dev);That'll work, too. Though, I find the notion of
__free()and a temporary variable__deva bit too cumbersome; it's used only for auto-clean stuff, so it could be somewhat anonymous.
No, it is not used only for auto-clean, it is also used for return value checking and it represents a reference on the original dev. It cannot be entirely anonymous because of the error checking part.
The point is that this is one statement instead of two and so it is arguably harder to mess up with.
Quoted 23 lines
But it's all about a matter of taste, and I'd follow what you and other guys suggest.
FWIW, there are lots of code doing like
pm_runtime_get_sync(dev); mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put(dev); return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put_autosuspend(dev); return 0;
Quoted 6 lines
and they can be converted nicely with
guard()once when PM runtime can be automatically unreferenced. With my proposed change, it would become like:pm_runtime_get_sync(dev); pm_runtime_auto_clean(dev);
For the case in which the pm_runtime_get_sync() return value is discarded, you could define a guard and do
guard(pm_runtime_get_sync)(dev);
here.
The case checking the return value is less straightforward.
Quoted 15 lines
guard(mutex)(&foo); .... return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; pm_runtime_auto_clean_autosuspend(dev); guard(mutex)(&foo); .... return 0;
Rafael J. Wysocki
maintainerQuoted 108 lines · Rafael J. Wysocki
Quoted 61 lines · Takashi Iwai
Quoted 58 lines · Rafael J. Wysocki
Hi,
Sorry for the delay.
Quoted 30 lines · Takashi Iwai
Quoted 27 lines · Takashi Iwai
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the
__free(pm_runtime_put)in commitbfa4477751e9, and there is a (single) user of it inpci-sysfs.c.Now I wanted to extend it to
pm_runtime_put_autosuspend()as:DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;that is similar as done in
pci-sysfs.c. So far, so good.But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
Quoted 7 lines
Quoted 2 lines
#define
pm_runtime_auto_clean(dev, var)\
struct device *var__free(pm_runtime_put)= (dev)It can be even simpler by assigning a temporary variable such as:
#define pm_runtime_auto_clean(dev) \ struct device *__pm_runtime_var ## __LINE__ __free(pm_runtime_put) = (dev)Well, if there's something like
struct device *pm_runtime_resume_and_get_dev(struct device *dev)
{ int ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ERR_PTR(ret); return dev; }It would be a matter of redefining the FREE to also take error pointers into account and you could do
Quoted 7 lines
Quoted 3 lines
struct device
*__dev__free(pm_runtim_put)=pm_runtime_resume_and_get_dev(dev); if (IS_ERR(__dev))return PTR_ERR(__dev);That'll work, too. Though, I find the notion of
__free()and a temporary variable__deva bit too cumbersome; it's used only for auto-clean stuff, so it could be somewhat anonymous.No, it is not used only for auto-clean, it is also used for return value checking and it represents a reference on the original dev. It cannot be entirely anonymous because of the error checking part.
The point is that this is one statement instead of two and so it is arguably harder to mess up with.
Quoted 23 lines
But it's all about a matter of taste, and I'd follow what you and other guys suggest.
FWIW, there are lots of code doing like
pm_runtime_get_sync(dev); mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put(dev); return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put_autosuspend(dev); return 0;Quoted 6 lines
and they can be converted nicely with
guard()once when PM runtime can be automatically unreferenced. With my proposed change, it would become like:pm_runtime_get_sync(dev); pm_runtime_auto_clean(dev);
Quoted 23 lines
For the case in which the
pm_runtime_get_sync()return value is discarded, you could define a guard and doguard(pm_runtime_get_sync)(dev);
here.
The case checking the return value is less straightforward.
Quoted 14 lines
guard(mutex)(&foo); .... return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; pm_runtime_auto_clean_autosuspend(dev); guard(mutex)(&foo); .... return 0;
I guess what I'm saying means basically something like this:
DEFINE_CLASS(pm_runtime_resume_and_get, struct device *,
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)
DEFINE_CLASS(pm_runtime_resume_and_get_auto, struct device *,
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put_autosuspend(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)
and analogously for pm_runtime_get_sync().
Rafael J. Wysocki
maintainerQuoted 137 lines · Rafael J. Wysocki
Quoted 108 lines · Rafael J. Wysocki
Quoted 61 lines · Takashi Iwai
Quoted 58 lines · Rafael J. Wysocki
Hi,
Sorry for the delay.
Quoted 30 lines · Takashi Iwai
Quoted 27 lines · Takashi Iwai
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the
__free(pm_runtime_put)in commitbfa4477751e9, and there is a (single) user of it inpci-sysfs.c.Now I wanted to extend it to
pm_runtime_put_autosuspend()as:DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;that is similar as done in
pci-sysfs.c. So far, so good.But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
Quoted 7 lines
Quoted 2 lines
#define
pm_runtime_auto_clean(dev, var)\
struct device *var__free(pm_runtime_put)= (dev)It can be even simpler by assigning a temporary variable such as:
#define pm_runtime_auto_clean(dev) \ struct device *__pm_runtime_var ## __LINE__ __free(pm_runtime_put) = (dev)Well, if there's something like
struct device *pm_runtime_resume_and_get_dev(struct device *dev)
{ int ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ERR_PTR(ret); return dev; }It would be a matter of redefining the FREE to also take error pointers into account and you could do
Quoted 7 lines
Quoted 3 lines
struct device
*__dev__free(pm_runtim_put)=pm_runtime_resume_and_get_dev(dev); if (IS_ERR(__dev))return PTR_ERR(__dev);That'll work, too. Though, I find the notion of
__free()and a temporary variable__deva bit too cumbersome; it's used only for auto-clean stuff, so it could be somewhat anonymous.No, it is not used only for auto-clean, it is also used for return value checking and it represents a reference on the original dev. It cannot be entirely anonymous because of the error checking part.
The point is that this is one statement instead of two and so it is arguably harder to mess up with.
Quoted 23 lines
But it's all about a matter of taste, and I'd follow what you and other guys suggest.
FWIW, there are lots of code doing like
pm_runtime_get_sync(dev); mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put(dev); return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put_autosuspend(dev); return 0;Quoted 6 lines
and they can be converted nicely with
guard()once when PM runtime can be automatically unreferenced. With my proposed change, it would become like:pm_runtime_get_sync(dev); pm_runtime_auto_clean(dev);Quoted 23 lines
For the case in which the
pm_runtime_get_sync()return value is discarded, you could define a guard and doguard(pm_runtime_get_sync)(dev);
here.
The case checking the return value is less straightforward.
Quoted 14 lines
guard(mutex)(&foo); .... return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; pm_runtime_auto_clean_autosuspend(dev); guard(mutex)(&foo); .... return 0;I guess what I'm saying means basically something like this:
DEFINE_CLASS(pm_runtime_resume_and_get, struct device *,
Quoted 8 lines
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)DEFINE_CLASS(pm_runtime_resume_and_get_auto, struct device *,
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put_autosuspend(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)and analogously for
pm_runtime_get_sync().
And it kind of makes sense either. Do
CLASS(pm_runtime_resume_and_get, active_dev)(dev); if (IS_ERR(active_dev))
return PTR_ERR(active_dev);
and now use active_dev for representing the device until it gets out of the scope.
Quoted 148 lines · Rafael J. Wysocki
Quoted 137 lines · Rafael J. Wysocki
Quoted 108 lines · Rafael J. Wysocki
Quoted 61 lines · Takashi Iwai
Quoted 58 lines · Rafael J. Wysocki
Hi,
Sorry for the delay.
Quoted 30 lines · Takashi Iwai
Quoted 27 lines · Takashi Iwai
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the
__free(pm_runtime_put)in commitbfa4477751e9, and there is a (single) user of it inpci-sysfs.c.Now I wanted to extend it to
pm_runtime_put_autosuspend()as:DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;that is similar as done in
pci-sysfs.c. So far, so good.But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
Quoted 7 lines
Quoted 2 lines
#define
pm_runtime_auto_clean(dev, var)\
struct device *var__free(pm_runtime_put)= (dev)It can be even simpler by assigning a temporary variable such as:
#define pm_runtime_auto_clean(dev) \ struct device *__pm_runtime_var ## __LINE__ __free(pm_runtime_put) = (dev)Well, if there's something like
struct device *pm_runtime_resume_and_get_dev(struct device *dev)
{ int ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ERR_PTR(ret); return dev; }It would be a matter of redefining the FREE to also take error pointers into account and you could do
Quoted 7 lines
Quoted 3 lines
struct device
*__dev__free(pm_runtim_put)=pm_runtime_resume_and_get_dev(dev); if (IS_ERR(__dev))return PTR_ERR(__dev);That'll work, too. Though, I find the notion of
__free()and a temporary variable__deva bit too cumbersome; it's used only for auto-clean stuff, so it could be somewhat anonymous.No, it is not used only for auto-clean, it is also used for return value checking and it represents a reference on the original dev. It cannot be entirely anonymous because of the error checking part.
The point is that this is one statement instead of two and so it is arguably harder to mess up with.
Quoted 23 lines
But it's all about a matter of taste, and I'd follow what you and other guys suggest.
FWIW, there are lots of code doing like
pm_runtime_get_sync(dev); mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put(dev); return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put_autosuspend(dev); return 0;Quoted 6 lines
and they can be converted nicely with
guard()once when PM runtime can be automatically unreferenced. With my proposed change, it would become like:pm_runtime_get_sync(dev); pm_runtime_auto_clean(dev);Quoted 23 lines
For the case in which the
pm_runtime_get_sync()return value is discarded, you could define a guard and doguard(pm_runtime_get_sync)(dev);
here.
The case checking the return value is less straightforward.
Quoted 14 lines
guard(mutex)(&foo); .... return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; pm_runtime_auto_clean_autosuspend(dev); guard(mutex)(&foo); .... return 0;I guess what I'm saying means basically something like this:
DEFINE_CLASS(pm_runtime_resume_and_get, struct device *,
Quoted 8 lines
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)DEFINE_CLASS(pm_runtime_resume_and_get_auto, struct device *,
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put_autosuspend(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)and analogously for
pm_runtime_get_sync().
Quoted 8 lines
And it kind of makes sense either. Do
CLASS(pm_runtime_resume_and_get, active_dev)(dev); if (IS_ERR(active_dev))return PTR_ERR(active_dev);and now use
active_devfor representing the device until it gets out of the scope.
Yes, that's what I thought of as an alternative, too, but I didn't consider using only pm_runtime_resume_and_get(). Actually by this action, we can also "clean up" the API usage at the same time to use a single recommended API function, which is a good thing.
That said, I like this way :)
It'd be nice if this change can go into 6.18, then I can put the driver cleanup works for 6.19. It's a bit late stage for 6.18, but this change is definitely safe and can't break, per se.
Thanks!
Takashi
Rafael J. Wysocki
maintainerQuoted 164 lines · CEST Takashi Iwai
Quoted 148 lines · Rafael J. Wysocki
Quoted 137 lines · Rafael J. Wysocki
Quoted 108 lines · Rafael J. Wysocki
Quoted 61 lines · Takashi Iwai
Quoted 58 lines · Rafael J. Wysocki
Hi,
Sorry for the delay.
Quoted 30 lines · Takashi Iwai
Quoted 27 lines · Takashi Iwai
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the
__free(pm_runtime_put)in commitbfa4477751e9, and there is a (single) user of it inpci-sysfs.c.Now I wanted to extend it to
pm_runtime_put_autosuspend()as:DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;that is similar as done in
pci-sysfs.c. So far, so good.But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
Quoted 7 lines
Quoted 2 lines
#define
pm_runtime_auto_clean(dev, var)\
struct device *var__free(pm_runtime_put)= (dev)It can be even simpler by assigning a temporary variable such as:
#define pm_runtime_auto_clean(dev) \ struct device *__pm_runtime_var ## __LINE__ __free(pm_runtime_put) = (dev)Well, if there's something like
struct device *pm_runtime_resume_and_get_dev(struct device *dev)
{ int ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ERR_PTR(ret); return dev; }It would be a matter of redefining the FREE to also take error pointers into account and you could do
Quoted 7 lines
Quoted 3 lines
struct device
*__dev__free(pm_runtim_put)=pm_runtime_resume_and_get_dev(dev); if (IS_ERR(__dev))return PTR_ERR(__dev);That'll work, too. Though, I find the notion of
__free()and a temporary variable__deva bit too cumbersome; it's used only for auto-clean stuff, so it could be somewhat anonymous.No, it is not used only for auto-clean, it is also used for return value checking and it represents a reference on the original dev. It cannot be entirely anonymous because of the error checking part.
The point is that this is one statement instead of two and so it is arguably harder to mess up with.
Quoted 23 lines
But it's all about a matter of taste, and I'd follow what you and other guys suggest.
FWIW, there are lots of code doing like
pm_runtime_get_sync(dev); mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put(dev); return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put_autosuspend(dev); return 0;Quoted 6 lines
and they can be converted nicely with
guard()once when PM runtime can be automatically unreferenced. With my proposed change, it would become like:pm_runtime_get_sync(dev); pm_runtime_auto_clean(dev);Quoted 23 lines
For the case in which the
pm_runtime_get_sync()return value is discarded, you could define a guard and doguard(pm_runtime_get_sync)(dev);
here.
The case checking the return value is less straightforward.
Quoted 14 lines
guard(mutex)(&foo); .... return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; pm_runtime_auto_clean_autosuspend(dev); guard(mutex)(&foo); .... return 0;I guess what I'm saying means basically something like this:
DEFINE_CLASS(pm_runtime_resume_and_get, struct device *,
Quoted 8 lines
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)DEFINE_CLASS(pm_runtime_resume_and_get_auto, struct device *,
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put_autosuspend(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)and analogously for
pm_runtime_get_sync().Quoted 8 lines
And it kind of makes sense either. Do
CLASS(pm_runtime_resume_and_get, active_dev)(dev); if (IS_ERR(active_dev))return PTR_ERR(active_dev);and now use
active_devfor representing the device until it gets out of the scope.Yes, that's what I thought of as an alternative, too, but I didn't consider using only
pm_runtime_resume_and_get(). Actually by this action, we can also "clean up" the API usage at the same time to use a single recommended API function, which is a good thing.
Quoted 5 lines
That said, I like this way :)
It'd be nice if this change can go into 6.18, then I can put the driver cleanup works for 6.19. It's a bit late stage for 6.18, but this change is definitely safe and can't break, per se.
OK, do you mean something like the patch below?
--- include/linux/pm_runtime.h | 43 +++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 43 insertions(+)
include/linux/pm_runtime.h +43 -0
@@ -533,6 +533,30 @@ static inline int pm_runtime_resume_and_ }/** +* pm_runtime_resume_and_get_dev - Resume device and bump up its usage counter. +* @dev: Target device. +* +* Resume @dev synchronously and if that is successful, increment its runtime +* PM usage counter. +* +* Return: +* * 0 if the runtime PM usage counter of @dev has been incremented. +* * Negative error code otherwise. +*/ +static inline struct device *pm_runtime_resume_and_get_dev (struct device *dev) +{ +int ret; + + ret =__pm_runtime_resume (dev,RPM_GET_PUT ); +if (ret <0 ) { +pm_runtime_put_noidle (dev); +return ERR_PTR (ret); + } + +return dev; +} + +/** * pm_runtime_put - Drop device usage counter and queue up "idle check" if 0. * @dev: Target device. *
@@ -606,6 +630,25 @@ static inline int pm_runtime_put_autosusreturn __pm_runtime_put_autosuspend (dev); } +/* +* The way to use the classes defined below is to define a class variable and +* use it going forward for representing the target device until it goes out of +* the scope. For example: +* +* CLASS(pm_runtime_resume_and_get, active_dev)(dev); +* if (IS_ERR(active_dev)) +* return PTR_ERR(active_dev); +* +* ... do something with active_dev (which is guaranteed to never suspend) ... +*/ +DEFINE_CLASS(pm_runtime_resume_and_get, struct device *, + if (!IS_ERR_OR_NULL(_T)) pm_runtime_put(_T), + pm_runtime_resume_and_get_dev(dev), struct device *dev) + +DEFINE_CLASS(pm_runtime_resume_and_get_auto, struct device *, + if (!IS_ERR_OR_NULL(_T)) pm_runtime_put_autosuspend(_T), + pm_runtime_resume_and_get_dev(dev), struct device *dev) +/** * pm_runtime_put_sync - Drop device usage counter and run "idle check" if 0. * @dev: Target device.
Quoted 166 lines · Rafael J. Wysocki
Quoted 164 lines · CEST Takashi Iwai
Quoted 148 lines · Rafael J. Wysocki
Quoted 137 lines · Rafael J. Wysocki
Quoted 108 lines · Rafael J. Wysocki
Quoted 61 lines · Takashi Iwai
Quoted 58 lines · Rafael J. Wysocki
Hi,
Sorry for the delay.
Quoted 30 lines · Takashi Iwai
Quoted 27 lines · Takashi Iwai
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the
__free(pm_runtime_put)in commitbfa4477751e9, and there is a (single) user of it inpci-sysfs.c.Now I wanted to extend it to
pm_runtime_put_autosuspend()as:DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;that is similar as done in
pci-sysfs.c. So far, so good.But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
Quoted 7 lines
Quoted 2 lines
#define
pm_runtime_auto_clean(dev, var)\
struct device *var__free(pm_runtime_put)= (dev)It can be even simpler by assigning a temporary variable such as:
#define pm_runtime_auto_clean(dev) \ struct device *__pm_runtime_var ## __LINE__ __free(pm_runtime_put) = (dev)Well, if there's something like
struct device *pm_runtime_resume_and_get_dev(struct device *dev)
{ int ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ERR_PTR(ret); return dev; }It would be a matter of redefining the FREE to also take error pointers into account and you could do
Quoted 7 lines
Quoted 3 lines
struct device
*__dev__free(pm_runtim_put)=pm_runtime_resume_and_get_dev(dev); if (IS_ERR(__dev))return PTR_ERR(__dev);That'll work, too. Though, I find the notion of
__free()and a temporary variable__deva bit too cumbersome; it's used only for auto-clean stuff, so it could be somewhat anonymous.No, it is not used only for auto-clean, it is also used for return value checking and it represents a reference on the original dev. It cannot be entirely anonymous because of the error checking part.
The point is that this is one statement instead of two and so it is arguably harder to mess up with.
Quoted 23 lines
But it's all about a matter of taste, and I'd follow what you and other guys suggest.
FWIW, there are lots of code doing like
pm_runtime_get_sync(dev); mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put(dev); return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put_autosuspend(dev); return 0;Quoted 6 lines
and they can be converted nicely with
guard()once when PM runtime can be automatically unreferenced. With my proposed change, it would become like:pm_runtime_get_sync(dev); pm_runtime_auto_clean(dev);Quoted 23 lines
For the case in which the
pm_runtime_get_sync()return value is discarded, you could define a guard and doguard(pm_runtime_get_sync)(dev);
here.
The case checking the return value is less straightforward.
Quoted 14 lines
guard(mutex)(&foo); .... return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; pm_runtime_auto_clean_autosuspend(dev); guard(mutex)(&foo); .... return 0;I guess what I'm saying means basically something like this:
DEFINE_CLASS(pm_runtime_resume_and_get, struct device *,
Quoted 8 lines
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)DEFINE_CLASS(pm_runtime_resume_and_get_auto, struct device *,
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put_autosuspend(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)and analogously for
pm_runtime_get_sync().Quoted 8 lines
And it kind of makes sense either. Do
CLASS(pm_runtime_resume_and_get, active_dev)(dev); if (IS_ERR(active_dev))return PTR_ERR(active_dev);and now use
active_devfor representing the device until it gets out of the scope.Yes, that's what I thought of as an alternative, too, but I didn't consider using only
pm_runtime_resume_and_get(). Actually by this action, we can also "clean up" the API usage at the same time to use a single recommended API function, which is a good thing.
Quoted 7 lines
Quoted 5 lines
That said, I like this way :)
It'd be nice if this change can go into 6.18, then I can put the driver cleanup works for 6.19. It's a bit late stage for 6.18, but this change is definitely safe and can't break, per se.
OK, do you mean something like the patch below?
Yes!
An easy follower is the patch like below.
(It's the only user of __free(pm_runtime_*) in linux-next as of now.)
thanks,
Takashi
-- 8< --
Subject: [PATCH] PCI: Use PM runtime class macro for the auto cleanup
The newly introduced class macro can simplify the code. Also, add the proper error handling for the PM runtime get, too.
--- drivers/pci/pci-sysfs.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/pci/pci-sysfs.c +3 -2
@@ -1475,8 +1475,9 @@ static ssize_t reset_method_store(struct device *dev,return count; } -pm_runtime_get_sync (dev); - struct device *pmdev __free(pm_runtime_put) = dev; +CLASS (pm_runtime_and_get, pmdev)(dev); +if (IS_ERR (pmdev)) +return PTR_ERR (pmdev);if (sysfs_streq (buf,"default" )) {pci_init_reset_methods (pdev);
Signature · 1 line
2.50.1
Quoted 175 lines · Takashi Iwai
Quoted 172 lines · Rafael J. Wysocki
Quoted 169 lines · CEST Takashi Iwai
Quoted 148 lines · Rafael J. Wysocki
Quoted 137 lines · Rafael J. Wysocki
Quoted 108 lines · Rafael J. Wysocki
Quoted 61 lines · Takashi Iwai
Quoted 58 lines · Rafael J. Wysocki
Hi,
Sorry for the delay.
Quoted 30 lines · Takashi Iwai
Quoted 27 lines · Takashi Iwai
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the
__free(pm_runtime_put)in commitbfa4477751e9, and there is a (single) user of it inpci-sysfs.c.Now I wanted to extend it to
pm_runtime_put_autosuspend()as:DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;that is similar as done in
pci-sysfs.c. So far, so good.But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
Quoted 7 lines
Quoted 2 lines
#define
pm_runtime_auto_clean(dev, var)\
struct device *var__free(pm_runtime_put)= (dev)It can be even simpler by assigning a temporary variable such as:
#define pm_runtime_auto_clean(dev) \ struct device *__pm_runtime_var ## __LINE__ __free(pm_runtime_put) = (dev)Well, if there's something like
struct device *pm_runtime_resume_and_get_dev(struct device *dev)
{ int ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ERR_PTR(ret); return dev; }It would be a matter of redefining the FREE to also take error pointers into account and you could do
Quoted 7 lines
Quoted 3 lines
struct device
*__dev__free(pm_runtim_put)=pm_runtime_resume_and_get_dev(dev); if (IS_ERR(__dev))return PTR_ERR(__dev);That'll work, too. Though, I find the notion of
__free()and a temporary variable__deva bit too cumbersome; it's used only for auto-clean stuff, so it could be somewhat anonymous.No, it is not used only for auto-clean, it is also used for return value checking and it represents a reference on the original dev. It cannot be entirely anonymous because of the error checking part.
The point is that this is one statement instead of two and so it is arguably harder to mess up with.
Quoted 23 lines
But it's all about a matter of taste, and I'd follow what you and other guys suggest.
FWIW, there are lots of code doing like
pm_runtime_get_sync(dev); mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put(dev); return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put_autosuspend(dev); return 0;Quoted 6 lines
and they can be converted nicely with
guard()once when PM runtime can be automatically unreferenced. With my proposed change, it would become like:pm_runtime_get_sync(dev); pm_runtime_auto_clean(dev);Quoted 23 lines
For the case in which the
pm_runtime_get_sync()return value is discarded, you could define a guard and doguard(pm_runtime_get_sync)(dev);
here.
The case checking the return value is less straightforward.
Quoted 14 lines
guard(mutex)(&foo); .... return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; pm_runtime_auto_clean_autosuspend(dev); guard(mutex)(&foo); .... return 0;I guess what I'm saying means basically something like this:
DEFINE_CLASS(pm_runtime_resume_and_get, struct device *,
Quoted 8 lines
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)DEFINE_CLASS(pm_runtime_resume_and_get_auto, struct device *,
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put_autosuspend(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)and analogously for
pm_runtime_get_sync().Quoted 8 lines
And it kind of makes sense either. Do
CLASS(pm_runtime_resume_and_get, active_dev)(dev); if (IS_ERR(active_dev))return PTR_ERR(active_dev);and now use
active_devfor representing the device until it gets out of the scope.Yes, that's what I thought of as an alternative, too, but I didn't consider using only
pm_runtime_resume_and_get(). Actually by this action, we can also "clean up" the API usage at the same time to use a single recommended API function, which is a good thing.That said, I like this way :)
It'd be nice if this change can go into 6.18, then I can put the driver cleanup works for 6.19. It's a bit late stage for 6.18, but this change is definitely safe and can't break, per se.
Quoted 5 lines
Quoted 1 line
OK, do you mean something like the patch below?
Yes!
An easy follower is the patch like below.
Err, I forgot to refresh before generating a patch. The proper one is below.
Takashi
-- 8< --
Subject: [PATCH] PCI: Use PM runtime class macro for the auto cleanup
The newly introduced class macro can simplify the code. Also, add the proper error handling for the PM runtime get, too.
--- drivers/pci/pci-sysfs.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/pci/pci-sysfs.c +3 -2
@@ -1475,8 +1475,9 @@ static ssize_t reset_method_store(struct device *dev,return count; } -pm_runtime_get_sync (dev); - struct device *pmdev __free(pm_runtime_put) = dev; +CLASS (pm_runtime_resume_and_get, pmdev)(dev); +if (IS_ERR (pmdev)) +return PTR_ERR (pmdev);if (sysfs_streq (buf,"default" )) {pci_init_reset_methods (pdev);
Signature · 1 line
2.50.1
Rafael J. Wysocki
maintainerQuoted 206 lines · Takashi Iwai
Quoted 175 lines · Takashi Iwai
Quoted 172 lines · Rafael J. Wysocki
Quoted 169 lines · CEST Takashi Iwai
Quoted 148 lines · Rafael J. Wysocki
Quoted 137 lines · Rafael J. Wysocki
Quoted 108 lines · Rafael J. Wysocki
Quoted 61 lines · Takashi Iwai
Quoted 58 lines · Rafael J. Wysocki
Hi,
Sorry for the delay.
Quoted 30 lines · Takashi Iwai
Quoted 27 lines · Takashi Iwai
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the
__free(pm_runtime_put)in commitbfa4477751e9, and there is a (single) user of it inpci-sysfs.c.Now I wanted to extend it to
pm_runtime_put_autosuspend()as:DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;that is similar as done in
pci-sysfs.c. So far, so good.But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
Quoted 7 lines
Quoted 2 lines
#define
pm_runtime_auto_clean(dev, var)\
struct device *var__free(pm_runtime_put)= (dev)It can be even simpler by assigning a temporary variable such as:
#define pm_runtime_auto_clean(dev) \ struct device *__pm_runtime_var ## __LINE__ __free(pm_runtime_put) = (dev)Well, if there's something like
struct device *pm_runtime_resume_and_get_dev(struct device *dev)
{ int ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ERR_PTR(ret); return dev; }It would be a matter of redefining the FREE to also take error pointers into account and you could do
Quoted 7 lines
Quoted 3 lines
struct device
*__dev__free(pm_runtim_put)=pm_runtime_resume_and_get_dev(dev); if (IS_ERR(__dev))return PTR_ERR(__dev);That'll work, too. Though, I find the notion of
__free()and a temporary variable__deva bit too cumbersome; it's used only for auto-clean stuff, so it could be somewhat anonymous.No, it is not used only for auto-clean, it is also used for return value checking and it represents a reference on the original dev. It cannot be entirely anonymous because of the error checking part.
The point is that this is one statement instead of two and so it is arguably harder to mess up with.
Quoted 23 lines
But it's all about a matter of taste, and I'd follow what you and other guys suggest.
FWIW, there are lots of code doing like
pm_runtime_get_sync(dev); mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put(dev); return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put_autosuspend(dev); return 0;Quoted 6 lines
and they can be converted nicely with
guard()once when PM runtime can be automatically unreferenced. With my proposed change, it would become like:pm_runtime_get_sync(dev); pm_runtime_auto_clean(dev);Quoted 23 lines
For the case in which the
pm_runtime_get_sync()return value is discarded, you could define a guard and doguard(pm_runtime_get_sync)(dev);
here.
The case checking the return value is less straightforward.
Quoted 14 lines
guard(mutex)(&foo); .... return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; pm_runtime_auto_clean_autosuspend(dev); guard(mutex)(&foo); .... return 0;I guess what I'm saying means basically something like this:
DEFINE_CLASS(pm_runtime_resume_and_get, struct device *,
Quoted 8 lines
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)DEFINE_CLASS(pm_runtime_resume_and_get_auto, struct device *,
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put_autosuspend(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)and analogously for
pm_runtime_get_sync().Quoted 8 lines
And it kind of makes sense either. Do
CLASS(pm_runtime_resume_and_get, active_dev)(dev); if (IS_ERR(active_dev))return PTR_ERR(active_dev);and now use
active_devfor representing the device until it gets out of the scope.Yes, that's what I thought of as an alternative, too, but I didn't consider using only
pm_runtime_resume_and_get(). Actually by this action, we can also "clean up" the API usage at the same time to use a single recommended API function, which is a good thing.That said, I like this way :)
It'd be nice if this change can go into 6.18, then I can put the driver cleanup works for 6.19. It's a bit late stage for 6.18, but this change is definitely safe and can't break, per se.
Quoted 5 lines
Quoted 1 line
OK, do you mean something like the patch below?
Yes!
An easy follower is the patch like below.
Err, I forgot to refresh before generating a patch. The proper one is below.
Takashi
-- 8< --
Subject: [PATCH] PCI: Use PM runtime class macro for the auto cleanupThe newly introduced class macro can simplify the code. Also, add the proper error handling for the PM runtime get, too.
Signed-off-by Takashi Iwai <[email protected]>--- drivers/pci/pci-sysfs.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)drivers/pci/pci-sysfs.c +0
-0
drivers/pci/pci-sysfs.c static ssize_t reset_method_store(struct device *dev, · line 1480
Quoted 8 lines
drivers/pci/pci-sysfs.c +3
-2@@ -1475,8 +1475,9 @@ static ssize_t reset_method_store(struct device *dev,return count; } -pm_runtime_get_sync (dev); - struct device *pmdev __free(pm_runtime_put) = dev; +CLASS (pm_runtime_resume_and_get, pmdev)(dev); +if (IS_ERR (pmdev)) +return PTR_ERR (pmdev);
This error will propagate to user space AFAICS and some of the values may be confusing in that respect, so it may be better to return -ENXIO here.
Quoted 4 lines
drivers/pci/pci-sysfs.c +0
-1if (sysfs_streq (buf,"default" )) {pci_init_reset_methods (pdev); --
Quoted 211 lines · Rafael J. Wysocki
Quoted 209 lines · Takashi Iwai
Quoted 175 lines · Takashi Iwai
Quoted 172 lines · Rafael J. Wysocki
Quoted 169 lines · CEST Takashi Iwai
Quoted 148 lines · Rafael J. Wysocki
Quoted 137 lines · Rafael J. Wysocki
Quoted 108 lines · Rafael J. Wysocki
Quoted 61 lines · Takashi Iwai
Quoted 58 lines · Rafael J. Wysocki
Hi,
Sorry for the delay.
Quoted 30 lines · Takashi Iwai
Quoted 27 lines · Takashi Iwai
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the
__free(pm_runtime_put)in commitbfa4477751e9, and there is a (single) user of it inpci-sysfs.c.Now I wanted to extend it to
pm_runtime_put_autosuspend()as:DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;that is similar as done in
pci-sysfs.c. So far, so good.But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
Quoted 7 lines
Quoted 2 lines
#define
pm_runtime_auto_clean(dev, var)\
struct device *var__free(pm_runtime_put)= (dev)It can be even simpler by assigning a temporary variable such as:
#define pm_runtime_auto_clean(dev) \ struct device *__pm_runtime_var ## __LINE__ __free(pm_runtime_put) = (dev)Well, if there's something like
struct device *pm_runtime_resume_and_get_dev(struct device *dev)
{ int ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ERR_PTR(ret); return dev; }It would be a matter of redefining the FREE to also take error pointers into account and you could do
Quoted 7 lines
Quoted 3 lines
struct device
*__dev__free(pm_runtim_put)=pm_runtime_resume_and_get_dev(dev); if (IS_ERR(__dev))return PTR_ERR(__dev);That'll work, too. Though, I find the notion of
__free()and a temporary variable__deva bit too cumbersome; it's used only for auto-clean stuff, so it could be somewhat anonymous.No, it is not used only for auto-clean, it is also used for return value checking and it represents a reference on the original dev. It cannot be entirely anonymous because of the error checking part.
The point is that this is one statement instead of two and so it is arguably harder to mess up with.
Quoted 23 lines
But it's all about a matter of taste, and I'd follow what you and other guys suggest.
FWIW, there are lots of code doing like
pm_runtime_get_sync(dev); mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put(dev); return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put_autosuspend(dev); return 0;Quoted 6 lines
and they can be converted nicely with
guard()once when PM runtime can be automatically unreferenced. With my proposed change, it would become like:pm_runtime_get_sync(dev); pm_runtime_auto_clean(dev);Quoted 23 lines
For the case in which the
pm_runtime_get_sync()return value is discarded, you could define a guard and doguard(pm_runtime_get_sync)(dev);
here.
The case checking the return value is less straightforward.
Quoted 14 lines
guard(mutex)(&foo); .... return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; pm_runtime_auto_clean_autosuspend(dev); guard(mutex)(&foo); .... return 0;I guess what I'm saying means basically something like this:
DEFINE_CLASS(pm_runtime_resume_and_get, struct device *,
Quoted 8 lines
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)DEFINE_CLASS(pm_runtime_resume_and_get_auto, struct device *,
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put_autosuspend(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)and analogously for
pm_runtime_get_sync().Quoted 8 lines
And it kind of makes sense either. Do
CLASS(pm_runtime_resume_and_get, active_dev)(dev); if (IS_ERR(active_dev))return PTR_ERR(active_dev);and now use
active_devfor representing the device until it gets out of the scope.Yes, that's what I thought of as an alternative, too, but I didn't consider using only
pm_runtime_resume_and_get(). Actually by this action, we can also "clean up" the API usage at the same time to use a single recommended API function, which is a good thing.That said, I like this way :)
It'd be nice if this change can go into 6.18, then I can put the driver cleanup works for 6.19. It's a bit late stage for 6.18, but this change is definitely safe and can't break, per se.
Quoted 5 lines
Quoted 1 line
OK, do you mean something like the patch below?
Yes!
An easy follower is the patch like below.
Err, I forgot to refresh before generating a patch. The proper one is below.
Takashi
-- 8< --
Subject: [PATCH] PCI: Use PM runtime class macro for the auto cleanupThe newly introduced class macro can simplify the code. Also, add the proper error handling for the PM runtime get, too.
Signed-off-by Takashi Iwai <[email protected]>--- drivers/pci/pci-sysfs.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)drivers/pci/pci-sysfs.c +0
-0@@ -1475,8 +1475,9 @@ static ssize_t reset_method_store(struct device *dev,return count; }
Quoted 9 lines
drivers/pci/pci-sysfs.c static ssize_t reset_method_store(struct device *dev, · line 1480
Quoted 5 lines
drivers/pci/pci-sysfs.c +3
-2-pm_runtime_get_sync (dev); - struct device *pmdev __free(pm_runtime_put) = dev; +CLASS (pm_runtime_resume_and_get, pmdev)(dev); +if (IS_ERR (pmdev)) +return PTR_ERR (pmdev);This error will propagate to user space AFAICS and some of the values may be confusing in that respect, so it may be better to return
-ENXIOhere.
Fair enough.
thanks,
Takashi
Rafael J. Wysocki
maintainerQuoted 171 lines · Takashi Iwai
Quoted 168 lines · Rafael J. Wysocki
Quoted 166 lines · CEST Takashi Iwai
Quoted 148 lines · Rafael J. Wysocki
Quoted 137 lines · Rafael J. Wysocki
Quoted 108 lines · Rafael J. Wysocki
Quoted 61 lines · Takashi Iwai
Quoted 58 lines · Rafael J. Wysocki
Hi,
Sorry for the delay.
Quoted 30 lines · Takashi Iwai
Quoted 27 lines · Takashi Iwai
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the
__free(pm_runtime_put)in commitbfa4477751e9, and there is a (single) user of it inpci-sysfs.c.Now I wanted to extend it to
pm_runtime_put_autosuspend()as:DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;that is similar as done in
pci-sysfs.c. So far, so good.But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
Quoted 7 lines
Quoted 2 lines
#define
pm_runtime_auto_clean(dev, var)\
struct device *var__free(pm_runtime_put)= (dev)It can be even simpler by assigning a temporary variable such as:
#define pm_runtime_auto_clean(dev) \ struct device *__pm_runtime_var ## __LINE__ __free(pm_runtime_put) = (dev)Well, if there's something like
struct device *pm_runtime_resume_and_get_dev(struct device *dev)
{ int ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ERR_PTR(ret); return dev; }It would be a matter of redefining the FREE to also take error pointers into account and you could do
Quoted 7 lines
Quoted 3 lines
struct device
*__dev__free(pm_runtim_put)=pm_runtime_resume_and_get_dev(dev); if (IS_ERR(__dev))return PTR_ERR(__dev);That'll work, too. Though, I find the notion of
__free()and a temporary variable__deva bit too cumbersome; it's used only for auto-clean stuff, so it could be somewhat anonymous.No, it is not used only for auto-clean, it is also used for return value checking and it represents a reference on the original dev. It cannot be entirely anonymous because of the error checking part.
The point is that this is one statement instead of two and so it is arguably harder to mess up with.
Quoted 23 lines
But it's all about a matter of taste, and I'd follow what you and other guys suggest.
FWIW, there are lots of code doing like
pm_runtime_get_sync(dev); mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put(dev); return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put_autosuspend(dev); return 0;Quoted 6 lines
and they can be converted nicely with
guard()once when PM runtime can be automatically unreferenced. With my proposed change, it would become like:pm_runtime_get_sync(dev); pm_runtime_auto_clean(dev);Quoted 23 lines
For the case in which the
pm_runtime_get_sync()return value is discarded, you could define a guard and doguard(pm_runtime_get_sync)(dev);
here.
The case checking the return value is less straightforward.
Quoted 14 lines
guard(mutex)(&foo); .... return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; pm_runtime_auto_clean_autosuspend(dev); guard(mutex)(&foo); .... return 0;I guess what I'm saying means basically something like this:
DEFINE_CLASS(pm_runtime_resume_and_get, struct device *,
Quoted 8 lines
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)DEFINE_CLASS(pm_runtime_resume_and_get_auto, struct device *,
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put_autosuspend(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)and analogously for
pm_runtime_get_sync().Quoted 8 lines
And it kind of makes sense either. Do
CLASS(pm_runtime_resume_and_get, active_dev)(dev); if (IS_ERR(active_dev))return PTR_ERR(active_dev);and now use
active_devfor representing the device until it gets out of the scope.Yes, that's what I thought of as an alternative, too, but I didn't consider using only
pm_runtime_resume_and_get(). Actually by this action, we can also "clean up" the API usage at the same time to use a single recommended API function, which is a good thing.That said, I like this way :)
Quoted 7 lines
Quoted 5 lines
Quoted 3 lines
It'd be nice if this change can go into 6.18, then I can put the driver cleanup works for 6.19. It's a bit late stage for 6.18, but this change is definitely safe and can't break, per se.
OK, do you mean something like the patch below?
Yes!
OK
Quoted 2 lines
An easy follower is the patch like below.
(It's the only user of__free(pm_runtime_*)in linux-next as of now.)
So the __free(pm_runtime_*) could be dropped after this patch I suppose?
In that case, let me send a series of 3 patches which will add the new class definitions, switch over PCI to using them (your patch), and drop the existing pm_runtime_put FREE.
Quoted 179 lines · Rafael J. Wysocki
Quoted 177 lines · Takashi Iwai
Quoted 173 lines · Rafael J. Wysocki
Quoted 164 lines · CEST Takashi Iwai
Quoted 148 lines · Rafael J. Wysocki
Quoted 137 lines · Rafael J. Wysocki
Quoted 108 lines · Rafael J. Wysocki
Quoted 61 lines · Takashi Iwai
Quoted 58 lines · Rafael J. Wysocki
Hi,
Sorry for the delay.
Quoted 30 lines · Takashi Iwai
Quoted 27 lines · Takashi Iwai
Hi,
while I worked on the code cleanups in the drivers with the recent auto-cleanup macros, I noticed that pm_runtime_get*() and _put*() can be also managed with the auto-cleanup gracefully, too. Actually we already defined the
__free(pm_runtime_put)in commitbfa4477751e9, and there is a (single) user of it inpci-sysfs.c.Now I wanted to extend it to
pm_runtime_put_autosuspend()as:DEFINE_FREE(pm_runtime_put_autosuspend, struct device *,
if (_T) pm_runtime_put_autosuspend(_T))Then one can use it like
ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ret; struct device *pmdev __free(pm_runtime_put_autosuspend) = dev;that is similar as done in
pci-sysfs.c. So far, so good.But, I find putting the line like above at each place a bit ugly. So I'm wondering whether it'd be better to introduce some helper macros, e.g.
Quoted 7 lines
Quoted 2 lines
#define
pm_runtime_auto_clean(dev, var)\
struct device *var__free(pm_runtime_put)= (dev)It can be even simpler by assigning a temporary variable such as:
#define pm_runtime_auto_clean(dev) \ struct device *__pm_runtime_var ## __LINE__ __free(pm_runtime_put) = (dev)Well, if there's something like
struct device *pm_runtime_resume_and_get_dev(struct device *dev)
{ int ret = pm_runtime_resume_and_get(dev); if (ret < 0) return ERR_PTR(ret); return dev; }It would be a matter of redefining the FREE to also take error pointers into account and you could do
Quoted 7 lines
Quoted 3 lines
struct device
*__dev__free(pm_runtim_put)=pm_runtime_resume_and_get_dev(dev); if (IS_ERR(__dev))return PTR_ERR(__dev);That'll work, too. Though, I find the notion of
__free()and a temporary variable__deva bit too cumbersome; it's used only for auto-clean stuff, so it could be somewhat anonymous.No, it is not used only for auto-clean, it is also used for return value checking and it represents a reference on the original dev. It cannot be entirely anonymous because of the error checking part.
The point is that this is one statement instead of two and so it is arguably harder to mess up with.
Quoted 23 lines
But it's all about a matter of taste, and I'd follow what you and other guys suggest.
FWIW, there are lots of code doing like
pm_runtime_get_sync(dev); mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put(dev); return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; mutex_lock(&foo); .... mutex_unlock(&foo); pm_runtime_put_autosuspend(dev); return 0;Quoted 6 lines
and they can be converted nicely with
guard()once when PM runtime can be automatically unreferenced. With my proposed change, it would become like:pm_runtime_get_sync(dev); pm_runtime_auto_clean(dev);Quoted 23 lines
For the case in which the
pm_runtime_get_sync()return value is discarded, you could define a guard and doguard(pm_runtime_get_sync)(dev);
here.
The case checking the return value is less straightforward.
Quoted 14 lines
guard(mutex)(&foo); .... return;or
ret = pm_runtime_resume_and_get(dev); if (ret) return ret; pm_runtime_auto_clean_autosuspend(dev); guard(mutex)(&foo); .... return 0;I guess what I'm saying means basically something like this:
DEFINE_CLASS(pm_runtime_resume_and_get, struct device *,
Quoted 8 lines
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)DEFINE_CLASS(pm_runtime_resume_and_get_auto, struct device *,
if (!IS_ERR_OR_NULL(_T)) pm_tuntime_put_autosuspend(_T),
pm_runtime_resume_and_get_dev(dev), struct device *dev)and analogously for
pm_runtime_get_sync().Quoted 8 lines
And it kind of makes sense either. Do
CLASS(pm_runtime_resume_and_get, active_dev)(dev); if (IS_ERR(active_dev))return PTR_ERR(active_dev);and now use
active_devfor representing the device until it gets out of the scope.Yes, that's what I thought of as an alternative, too, but I didn't consider using only
pm_runtime_resume_and_get(). Actually by this action, we can also "clean up" the API usage at the same time to use a single recommended API function, which is a good thing.Quoted 5 lines
That said, I like this way :)
It'd be nice if this change can go into 6.18, then I can put the driver cleanup works for 6.19. It's a bit late stage for 6.18, but this change is definitely safe and can't break, per se.
OK, do you mean something like the patch below?
Quoted 8 lines
Quoted 1 line
Yes!
OK
Quoted 2 lines
An easy follower is the patch like below.
(It's the only user of__free(pm_runtime_*)in linux-next as of now.)So the
__free(pm_runtime_*)could be dropped after this patch I suppose?
Yes, for now it seems so. It was the only user as far as I can see in linux-next.
Quoted 3 lines
In that case, let me send a series of 3 patches which will add the new class definitions, switch over PCI to using them (your patch), and drop the existing
pm_runtime_putFREE.
OK, will do that.
Takashi