reset: Make optional stuff optional for all users
PATCH 1/1 reset: Make optional stuff optional for all users
There is Device Tree oriented check for optional resource. Of course it will fail on non-DT platforms.
Remove this check to make things optional for all users.
---
The reset framework is too Device Tree oriented, and who knows what the logic was behind the commit which introduced devm_reset_*() functions without thinking out of the DT box.
This commit fixes almost all Intel newest boards that have no legacy UART since UART driver started using this DT-centric framework.
drivers/reset/core.c | 3 --- 1 file changed, 3 deletions(-)
drivers/reset/core.c +0 -3
@@ -331,9 +331,6 @@ struct reset_control *__of_reset_control_get(struct device_node *node,int rstc_id;int ret; -if (!node) -return ERR_PTR (-EINVAL ); -if (id) { index =of_property_match_string (node,"reset-names" , id);
Signature · 1 line
2.11.0
Philipp Zabel
maintainerHi Andy,
thank you for the patch.
Quoted 5 lines · Andy Shevchenko
There is Device Tree oriented check for optional resource. Of course it will fail on non-DT platforms.
Remove this check to make things optional for all users.
Quoted 8 lines
Fixes bb475230b8e5 ("reset: make optional functions really optional")Cc Ramiro Oliveira <[email protected]>Signed-off-by Andy Shevchenko <[email protected]>---
The reset framework is too Device Tree oriented, and who knows what the logic was behind the commit which introduced devm_reset_*() functions without thinking out of the DT box.
At the time, there was nothing outside of the box to describe reset lines, and as far as I am aware there still isn't, so the devm_reset_* should behave as if the reset line is not specified in the non-DT case. Returning -EINVAL was reasonable in that case, before the API was changed to describe unavailable, optional reset controls as rstc = NULL.
Quoted 2 lines
This commit fixes almost all Intel newest boards that have no legacy UART since UART driver started using this DT-centric framework.
Is this is about 8250_dw.c? Unfortunately it sometimes takes a little while for me to get updated on the big picture, as I only get the actual reset driver and framework patches in my inbox. Usually I only see the reset consumer changes when I actively look for them.
But the fault in this case was is with me not considering all possible code paths influenced by commit bb475230b8e5, in the configurations that I can't test myself.
drivers/reset/core.c struct reset_control *__of_reset_control_get(struct device_node *node,
Quoted 8 lines
drivers/reset/core.c | 3 --- 1 file changed, 3 deletions(-)drivers/reset/core.c +0
-0@@ -331,9 +331,6 @@ struct reset_control *__of_reset_control_get(struct device_node *node,
Ideally, __of_reset_control_get would not be called at all in the non-DT case. I'll change that in the next round, but for now I'd prefer a small fix in place.
drivers/reset/core.c struct reset_control *__of_reset_control_get(struct device_node *node, · line 336
Quoted 6 lines
drivers/reset/core.c +0
-3int rstc_id;int ret; -if (!node) -return ERR_PTR (-EINVAL ); -
This should be
if (!node) return optional ? NULL : ERR_PTR(-EINVAL);
instead. Can you confirm this works for Intel boards with DW UART? I can fix it up when applying if you agree.
drivers/reset/core.c struct reset_control *__of_reset_control_get(struct device_node *node, · line 336
Quoted 3 lines
drivers/reset/core.c +0
-0if (id) { index =of_property_match_string (node,"reset-names" , id);
regards
Philipp
Quoted 20 lines · Philipp Zabel
Hi Andy,
thank you for the patch.
Quoted 6 lines · Andy Shevchenko
There is Device Tree oriented check for optional resource. Of course it will fail on non-DT platforms.
Remove this check to make things optional for all users.
Fixes bb475230b8e5 ("reset: make optional functions reallyQuoted 8 lines
optional")
Cc Ramiro Oliveira <[email protected]>Signed-off-by Andy Shevchenko <[email protected]>---
The reset framework is too Device Tree oriented, and who knows what the logic was behind the commit which introduced devm_reset_*() functions without thinking out of the DT box.
Quoted 8 lines
At the time, there was nothing outside of the box to describe reset lines, and as far as I am aware there still isn't, so the devm_reset_* should behave as if the reset line is not specified in the non-DT case.
Returning
-EINVALwas reasonable in that case, before the API was changed to describe unavailable, optional reset controls as rstc = NULL.
Fair enough.
Quoted 10 lines
Quoted 4 lines
This commit fixes almost all Intel newest boards that have no legacy
UART since UART driver started using this DT-centric framework.Is this is about
8250_dw.c? Unfortunately it sometimes takes a little while for me to get updated on the big picture, as I only get the actual
reset driver and framework patches in my inbox. Usually I only see the reset consumer changes when I actively look for them.
Yes, 8250_dw.c in this case.
Quoted 5 lines
But the fault in this case was is with me not considering all possible code paths influenced by commit
bb475230b8e5, in the configurations that
I can't test myself.
Understood.
Quoted 26 lines
drivers/reset/core.c struct reset_control · line 331
Quoted 10 lines
drivers/reset/core.c | 3 ---
1 file changed, 3 deletions(-)drivers/reset/core.c +0
-0@@ -331,9 +331,6 @@ struct reset_control *__of_reset_control_get(struct device_node *node,Ideally,
__of_reset_control_getwould not be called at all in the non- DT
case. I'll change that in the next round, but for now I'd prefer a small
fix in place.drivers/reset/core.c struct reset_control *__of_reset_control_get(struct device_node *node, · line 336
Quoted 6 lines
drivers/reset/core.c +0
-0int rstc_id; int ret;drivers/reset/core.c +0
-3-if (!node) -return ERR_PTR (-EINVAL ); -This should be
Quoted 6 lines
if (!node) return optional ? NULL : ERR_PTR(-EINVAL);instead. Can you confirm this works for Intel boards with DW UART? I can
fix it up when applying if you agree.
I don't think it worth to change. I specifically checked all of_* calls in that function and they cope pretty nice with node == NULL.
So, I rather to go with my initial change.
Thanks for review!
Signature · 2 lines
Andy Shevchenko <[email protected]> Intel Finland Oy
Quoted 19 lines · Andy Shevchenko
Quoted 11 lines · Philipp Zabel
drivers/reset/core.c struct reset_control *__of_reset_control_get(struct device_node *node, · line 336
Quoted 8 lines
int
rstc_id;
int ret;drivers/reset/core.c +0
-3-if (!node) -return ERR_PTR (-EINVAL ); -This should be
Quoted 6 lines
if (!node) return optional ? NULL : ERR_PTR(-EINVAL);instead. Can you confirm this works for Intel boards with DW UART? I can
fix it up when applying if you agree.
Quoted 6 lines
I don't think it worth to change. I specifically checked all of_* calls
in that function and they cope pretty nice with node == NULL.So, I rather to go with my initial change.
Hit Enter before closing another thought.
When you come with solution where this __of_reset_control_get() will be called only for node != NULL case you will not need that check either.
So, I would go my solution because of two benefits:
- it fixes bug
- if will not bring ping-ponging code
Quoted 2 lines
Thanks for review!
Signature · 2 lines
Andy Shevchenko <[email protected]> Intel Finland Oy - BIC 0357606-4 - Westendinkatu 7, 02160 Espoo
Philipp Zabel
maintainerQuoted 16 lines · Andriy
Quoted 15 lines · Andy Shevchenko
Quoted 14 lines · Philipp Zabel
drivers/reset/core.c struct reset_control *__of_reset_control_get(struct device_node *node, · line 336
Quoted 8 lines
drivers/reset/core.c +0
-3int rstc_id;int ret; -if (!node) -return ERR_PTR (-EINVAL ); -This should be
if (!node) return optional ? NULL : ERR_PTR(-EINVAL);
Quoted 7 lines
Quoted 7 lines
Quoted 3 lines
instead. Can you confirm this works for Intel boards with DW UART? I can
fix it up when applying if you agree.I don't think it worth to change. I specifically checked all of_* calls
in that function and they cope pretty nice with node == NULL.
__of_reset_control_get called with id != NULL calls of_property_match_string first, which then returns -EINVAL if node == NULL, which makes __of_reset_control_get return NULL if optional or -ENOENT otherwise, even though the correct return value would be -EINVAL in the DT case.
__of_reset_control_get called with id == NULL calls of_parse_phandle_with_args first, which calls __of_parse_phandle_with_args, which returns an undefined value if np == NULL, as far as I can tell:of_for_each_phandle first calls of_phandle_iterator_init, which, when called with np == NULL clears the iterator structure returns -ENOENT. The return value is ignored in the of_for_each_phandle macro, and of_phandle_iterator_next is then called and returns -ENOENT because it->cur == NULL, ending the loop without ever assigning a value to rc. __of_parse_phandle_with_args then returns the uninitialized value.
The point being, instead of having to regularly forage through a number of of_ API functions to make sure my expectations are still met, I'd prefer to keep the check in place.
Quoted 8 lines
Quoted 3 lines
So, I rather to go with my initial change.
Hit Enter before closing another thought.
When you come with solution where this
__of_reset_control_get()will be called only for node != NULL case you will not need that check either.
__of_reset_control_get is public API (via of_reset_control_get), so I can't guarantee node != NULL even in the DT case.
Quoted 2 lines
So, I would go my solution because of two benefits:
- it fixes bug
True.
Quoted 1 line
- if will not bring ping-ponging code
Unfortunately not.
regards
Philipp
Quoted 25 lines · Philipp Zabel
Quoted 15 lines · Andriy
Quoted 14 lines · Andy Shevchenko
Quoted 13 lines · Philipp Zabel
drivers/reset/core.c struct reset_control *__of_reset_control_get(struct device_node *node, · line 336
Quoted 7 lines
int
rstc_id;
int ret;drivers/reset/core.c +0
-3-if (!node) -return ERR_PTR (-EINVAL ); -This should be
if (!node) return optional ? NULL : ERR_PTR(-EINVAL);Quoted 8 lines
Quoted 8 lines
Quoted 4 lines
instead. Can you confirm this works for Intel boards with DW UART? I
can
fix it up when applying if you agree.I don't think it worth to change. I specifically checked all of_* calls
in that function and they cope pretty nice with node == NULL.
Quoted 6 lines
__of_reset_control_getcalled with id != NULL callsof_property_match_stringfirst, which then returns-EINVALif node == NULL, which makes__of_reset_control_getreturn NULL if optional
or-ENOENTotherwise, even though the correct return value would be-EINVALin the DT case.
Error handling mess as usual. :-)
Quoted 11 lines
__of_reset_control_getcalled with id == NULL callsof_parse_phandle_with_argsfirst, which calls__of_parse_phandle_with_args, which returns an undefined value if np == NULL, as far as I can tell:of_for_each_phandlefirst callsof_phandle_iterator_init, which, when called with np == NULL clears the iterator structure returns-ENOENT. The return value is ignored in theof_for_each_phandlemacro, andof_phandle_iterator_nextis then called and returns-ENOENTbecauseit->cur== NULL, ending the loop without ever assigning a value to rc.__of_parse_phandle_with_argsthen returns the uninitialized value.
It returns -ENOENT. Error value is kept in function variable rc.
Quoted 5 lines
The point being, instead of having to regularly forage through a number
ofof_API functions to make sure my expectations are still met, I'd prefer to keep the check in place.
I would not insist, I already shared my view on this.
I really don't like ping-ponging of the code. Perhaps you would fix it once for the best now?
Quoted 7 lines
Quoted 7 lines
Quoted 4 lines
So, I rather to go with my initial change.
Hit Enter before closing another thought.
Quoted 7 lines
Quoted 4 lines
When you come with solution where this
__of_reset_control_get()will be
called only for node != NULL case you will not need that check either.
__of_reset_control_getis public API (viaof_reset_control_get), so I can't guarantee node != NULL even in the DT case.
Yes, and that's why other callees will return some error codes there.
Quoted 9 lines
Quoted 3 lines
So, I would go my solution because of two benefits:
- it fixes bugTrue.
Quoted 1 line
- if will not bring ping-ponging code
Unfortunately not.
Fortunately yes, if someone will fix DT error code mess in the first place.
Signature · 2 lines
Andy Shevchenko <[email protected]> Intel Finland Oy
Philipp Zabel
maintainerQuoted 47 lines · Andy Shevchenko
Quoted 25 lines · Philipp Zabel
Quoted 15 lines · Andriy
Quoted 14 lines · Andy Shevchenko
Quoted 13 lines · Philipp Zabel
Review
Quoted 7 lines
(quoted diff) +0
-3int rstc_id; int ret; - if (!node) - return ERR_PTR(-EINVAL); -This should be
if (!node) return optional ? NULL : ERR_PTR(-EINVAL);Quoted 8 lines
Quoted 8 lines
Quoted 4 lines
instead. Can you confirm this works for Intel boards with DW UART? I
can
fix it up when applying if you agree.I don't think it worth to change. I specifically checked all of_* calls
in that function and they cope pretty nice with node == NULL.Quoted 6 lines
__of_reset_control_getcalled with id != NULL callsof_property_match_stringfirst, which then returns-EINVALif node == NULL, which makes__of_reset_control_getreturn NULL if optional
or-ENOENTotherwise, even though the correct return value would be-EINVALin the DT case.Error handling mess as usual. :-)
Quoted 11 lines
__of_reset_control_getcalled with id == NULL callsof_parse_phandle_with_argsfirst, which calls__of_parse_phandle_with_args, which returns an undefined value if np == NULL, as far as I can tell:of_for_each_phandlefirst callsof_phandle_iterator_init, which, when called with np == NULL clears the iterator structure returns-ENOENT. The return value is ignored in theof_for_each_phandlemacro, andof_phandle_iterator_nextis then called and returns-ENOENTbecauseit->cur== NULL, ending the loop without ever assigning a value to rc.__of_parse_phandle_with_argsthen returns the uninitialized value.
Quoted 1 line
It returns
-ENOENT. Error value is kept in function variable rc.
Thanks for the correction. That's still the wrong error code IMHO, but it certainly makes your patch unproblematic.
Quoted 6 lines
Quoted 4 lines
The point being, instead of having to regularly forage through a number
ofof_API functions to make sure my expectations are still met, I'd prefer to keep the check in place.I would not insist, I already shared my view on this.
Appreciated. With a correct understanding of the current of_* error return paths it makes a lot more sense.
Quoted 2 lines
I really don't like ping-ponging of the code. Perhaps you would fix it once for the best now?
How about:
----------8<----------
Quoted 1 line
From
75205aeaf52cef5c62710630cceb871468175c12Mon Sep 17 00:00:00 2001
From: Philipp Zabel <[email protected]>
Date: Fri, 5 Feb 2016 13:41:39 +0100
Subject: [PATCH] reset: add exported __reset_control_get, return NULL if
optional
Rename the internal __reset_control_get/put functions to __reset_control_get/put_internal and add an exported __reset_control_get equivalent to __of_reset_control_get that takes a struct device parameter.
This avoids the confusing call to __of_reset_control_get in the non-DT case and fixes the devm_reset_control_get_optional function to return NULL if RESET_CONTROLLER is enabled but dev->of_node == NULL.
--- drivers/reset/core.c | 22 ++++++++++++++++------ include/linux/reset.h | 22 ++++++++++++++-------- 2 files changed, 30 insertions(+), 14 deletions(-)
drivers/reset/core.c +16 -6
@@ -275,7 +275,7 @@ int reset_control_status(struct reset_control *rstc) }EXPORT_SYMBOL_GPL (reset_control_status); -static struct reset_control *__reset_control_get ( +static struct reset_control *__reset_control_get_internal (struct reset_controller_dev *rcdev,unsigned int index,bool shared) {
@@ -308,7 +308,7 @@ static struct reset_control *__reset_control_get(return rstc; } -static void __reset_control_put (struct reset_control *rstc) +static void __reset_control_put_internal (struct reset_control *rstc) {lockdep_assert_held (&reset_list_mutex);
@@ -377,7 +377,7 @@ struct reset_control *__of_reset_control_get(struct device_node *node, }/* reset_list_mutex also protects the rcdev's reset_control list */ - rstc =__reset_control_get (rcdev, rstc_id, shared); + rstc =__reset_control_get_internal (rcdev, rstc_id, shared);mutex_unlock (&reset_list_mutex);
@@ -385,6 +385,17 @@ struct reset_control *__of_reset_control_get(struct device_node *node, }EXPORT_SYMBOL_GPL (__of_reset_control_get); +struct reset_control *__reset_control_get (struct device *dev,const char *id, +int index,bool shared,bool optional) +{ +if (dev->of_node) +return __of_reset_control_get (dev->of_node, id, index, shared, + optional); +else +return optional ?NULL :ERR_PTR (-EINVAL ); +} +EXPORT_SYMBOL_GPL (__reset_control_get); +/** * reset_control_put - free the reset controller * @rstc: reset controller
@@ -396,7 +407,7 @@ void reset_control_put(struct reset_control *rstc)return ;mutex_lock (&reset_list_mutex); -__reset_control_put (rstc); +__reset_control_put_internal (rstc);mutex_unlock (&reset_list_mutex); }EXPORT_SYMBOL_GPL (reset_control_put);
@@ -417,8 +428,7 @@ struct reset_control *__devm_reset_control_get(struct device *dev,if (!ptr)return ERR_PTR (-ENOMEM ); - rstc =__of_reset_control_get (dev ? dev->of_node :NULL , - id, index, shared, optional); + rstc =__reset_control_get (dev, id, index, shared, optional);if (!IS_ERR (rstc)) { *ptr = rstc;devres_add (dev, ptr);
include/linux/reset.h +14 -8
@@ -15,6 +15,9 @@ int reset_control_status(struct reset_control *rstc);struct reset_control *__of_reset_control_get (struct device_node *node,const char *id,int index,bool shared,bool optional); +struct reset_control *__reset_control_get (struct device *dev,const char *id, +int index,bool shared, +bool optional);void reset_control_put (struct reset_control *rstc); struct reset_control *__devm_reset_control_get(struct device *dev, const char *id, int index, bool shared,
@@ -72,6 +75,13 @@ static inline struct reset_control *__of_reset_control_get(return optional ?NULL :ERR_PTR (-ENOTSUPP ); } +static inline struct reset_control *__reset_control_get ( +struct device *dev,const char *id, +int index,bool shared,bool optional) +{ +return optional ?NULL :ERR_PTR (-ENOTSUPP ); +} +static inline struct reset_control *__devm_reset_control_get (struct device *dev,const char *id,int index,bool shared,bool optional)
@@ -102,8 +112,7 @@ __must_check reset_control_get_exclusive(struct device *dev, const char *id)#ifndef CONFIG_RESET_CONTROLLER WARN_ON (1 );#endif -return __of_reset_control_get (dev ? dev->of_node :NULL , id,0 , false, - false); +return __reset_control_get (dev, id,0 , false, false); }/**
@@ -131,22 +140,19 @@ __must_check reset_control_get_exclusive(struct device *dev, const char *id)static inline struct reset_control *reset_control_get_shared (struct device *dev,const char *id) { -return __of_reset_control_get (dev ? dev->of_node :NULL , id,0 , true, - false); +return __reset_control_get (dev, id,0 , true, false); }static inline struct reset_control *reset_control_get_optional_exclusive (struct device *dev,const char *id) { -return __of_reset_control_get (dev ? dev->of_node :NULL , id,0 , false, - true); +return __reset_control_get (dev, id,0 , false, true); }static inline struct reset_control *reset_control_get_optional_shared (struct device *dev,const char *id) { -return __of_reset_control_get (dev ? dev->of_node :NULL , id,0 , true, - true); +return __reset_control_get (dev, id,0 , true, true); }/**
Signature · 10 lines
2.11.0 ---------->8---------- That should make both the __reset_control_get stub and implementation return NULL if optional && (dev->of_node == NULL), and stop calling __of_reset_control_get in the non-DT case, while maintaining the -EINVAL error return value if of_reset_control_get* is called with node == NULL. regards Philipp
Quoted 9 lines · Philipp Zabel
Quoted 6 lines · Andy Shevchenko
Quoted 2 lines · Philipp Zabel
Quoted 1 line · Andriy
On Mon, 2017-04-03 at 17:31 +0300, Andy Shevchenko wrote:
I really don't like ping-ponging of the code. Perhaps you would fix it once for the best now?
How about:
Below looks fine to me, one nit still and I would like to test it tomorrow. Thanks for taking care.
drivers/reset/core.c struct reset_control *__of_reset_control_get(struct device_node *node, · line 398
Quoted 6 lines
drivers/reset/core.c +6
-0+struct reset_control *__reset_control_get (struct device *dev,const char *id, +int index,bool shared,bool optional) +{ +if (dev->of_node) + return __of_reset_control_get(dev->of_node, id, index, shared, +
optional);
Quoted 1 line
+ else
Redundant.
Quoted 3 lines
drivers/reset/core.c +3
-0+return optional ?NULL :ERR_PTR (-EINVAL ); +} +EXPORT_SYMBOL_GPL (__reset_control_get);
Signature · 2 lines
With Best Regards, Andy Shevchenko