gpio: sim: improve the usage of __free()
v3 cover gpio: sim: improve the usage of __free()
From: Bartosz Golaszewski <[email protected]>
Hi Linus et al,
I'm sorry for the noise but v2 was incorrect, this is a fixed version.
As discussed here's an improved fix for the invalid usage of __free() in gpio-sim. I based it on your "maybe-sane" suggestion but unfortunately it missed a couple details that make it impossible to avoid a conditional initialization of the managed pointer without repeating the call to fwnode_create_software_node().
What we're doing here is: we're creating the string array for the standard "gpio-line-names" device property. It can look like this:
{ "foo", "bar", "baz" }In which case lines 0, 1 and 2 are named but it can also look like this:
{ "foo", NULL, NULL, "bar", NULL, "baz" }Where only lines 0, 3 and 5 have assigned names.
So the has_line_names boolean set when encountering the first line with a name is there for a reason, namely: it's possible that only the line at offset 0 will have a name, leaving max_offset at 0 but we still need to create an array of size 1 in this case.
If the array is created and filled, then it needs to live until a deep copy is completed in fwnode_create_software_node() so it has to be defined at the top of the function.
I think this still results in clearer code then if we called return fwnode_create_software_node(); twice with the same arguments.
I also changed the naming to reflect the purpose of the array: it can be sparse so it's not really "number of lines", it's the "size of the array holding the names". The array can be of size 10 but we can only have 3 named lines.
To atone for the above, I've added a second patch which changes the other instance of __free() in this driver to be initialized in place.
If this is alright for you, please consider applying it directly to your tree for v6.6-rc2.
Best regards,
Bartosz Golaszewski
v2 -> v3:
- restore the offset out-of-bounds checks
v1 -> v2:
- split the line name setting into two parts
- add a patch improving the second instance of using __free()
Bartosz Golaszewski (2):
gpio: sim: fix an invalid __free() usage
gpio: sim: initialize a managed pointer when declaring it
drivers/gpio/gpio-sim.c | 68 +++++++++++++++++++---------------------- 1 file changed, 32 insertions(+), 36 deletions(-)
Signature · 1 line
2.39.2
PATCH v3 1/2 gpio: sim: fix an invalid __free() usage
From: Bartosz Golaszewski <[email protected]>
gpio_sim_make_line_names() returns NULL or ERR_PTR() so we must not use __free(kfree) on the returned address. Split this function into two, one that determines the size of the "gpio-line-names" array to allocate and one that actually sets the names at correct offsets. The allocation and assignment of the managed pointer happens in between.
--- drivers/gpio/gpio-sim.c | 64 +++++++++++++++++++---------------------- 1 file changed, 30 insertions(+), 34 deletions(-)
drivers/gpio/gpio-sim.c +30 -34
@@ -21,6 +21,7 @@#include <linux/irq.h> #include <linux/irq_sim.h> #include <linux/list.h> +#include <linux/minmax.h> #include <linux/mod_devicetable.h> #include <linux/module.h> #include <linux/mutex.h>
@@ -718,52 +719,42 @@ gpio_sim_device_config_live_show(struct config_item *item, char *page)return sprintf (page,"%c\n" , live ?'1' :'0' ); } -static char **gpio_sim_make_line_names (struct gpio_sim_bank *bank, -unsigned int *line_names_size) +static unsigned int gpio_sim_get_line_names_size (struct gpio_sim_bank *bank) {unsigned int max_offset =0 ;bool has_line_names = false;struct gpio_sim_line *line; -char **line_names;list_for_each_entry (line, &bank->line_list, siblings) { -if (line->offset >= bank->num_lines) +if (!line->name || (line->offset >= bank->num_lines))continue ; -if (line->name) { -if (line->offset > max_offset) - max_offset = line->offset; - -/* -* max_offset can stay at 0 so it's not an indicator -* of whether line names were configured at all. -*/ - has_line_names = true; - } + has_line_names = true; + max_offset =max (line->offset, max_offset); } -if (!has_line_names) -/* -* This is not an error - NULL means, there are no line -* names configured. -*/ -return NULL ; +/* +* It's possible that only the line at offset 0 has a name in which +* case max_offset will be 0 but we still want to allocate an array +* of size 1. +*/ +if (has_line_names) + max_offset++; - *line_names_size = max_offset +1 ; +return max_offset; +} - line_names =kcalloc (*line_names_size,sizeof (*line_names),GFP_KERNEL ); -if (!line_names) -return ERR_PTR (-ENOMEM ); +static void +gpio_sim_set_line_names (struct gpio_sim_bank *bank,char **line_names) +{ +struct gpio_sim_line *line;list_for_each_entry (line, &bank->line_list, siblings) { -if (line->offset >= bank->num_lines) +if (!line->name || (line->offset >= bank->num_lines))continue ; -if (line->name && (line->offset <= max_offset)) - line_names[line->offset] = line->name; + line_names[line->offset] = line->name; } - -return line_names; }static void gpio_sim_remove_hogs (struct gpio_sim_device *dev)
@@ -867,7 +858,7 @@ gpio_sim_make_bank_swnode(struct gpio_sim_bank *bank, struct fwnode_handle *parent) {struct property_entry properties[GPIO_SIM_PROP_MAX ]; -unsigned int prop_idx =0 , line_names_size =0 ; +unsigned int prop_idx =0 , line_names_size; char **line_names __free(kfree) = NULL;memset (properties,0 ,sizeof (properties));
@@ -878,14 +869,19 @@ gpio_sim_make_bank_swnode(struct gpio_sim_bank *bank, properties[prop_idx++] =PROPERTY_ENTRY_STRING ("gpio-sim,label" , bank->label); - line_names =gpio_sim_make_line_names (bank, &line_names_size); -if (IS_ERR (line_names)) -return ERR_CAST (line_names); + line_names_size =gpio_sim_get_line_names_size (bank); +if (line_names_size) { + line_names =kcalloc (line_names_size,sizeof (*line_names), +GFP_KERNEL ); +if (!line_names) +return ERR_PTR (-ENOMEM ); + +gpio_sim_set_line_names (bank, line_names); -if (line_names) properties[prop_idx++] =PROPERTY_ENTRY_STRING_ARRAY_LEN ("gpio-line-names" , line_names, line_names_size); + }return fwnode_create_software_node (properties, parent); }
Signature · 1 line
2.39.2
PATCH v3 2/2 gpio: sim: initialize a managed pointer when declaring it
In reply to Bartosz GolaszewskiFrom: Bartosz Golaszewski <[email protected]>
Variables managed with __free() should typically be initialized where they are declared so that the __free() callback is paired with its counterpart resource allocator. Fix the second instance of using __free() in gpio-sim to follow this pattern.
--- drivers/gpio/gpio-sim.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpio/gpio-sim.c +2 -2
@@ -1481,10 +1481,10 @@ static const struct config_item_type gpio_sim_device_config_group_type = {static struct config_group *gpio_sim_config_make_device_group (struct config_group *group,const char *name) { - struct gpio_sim_device *dev __free(kfree) = NULL;int id; - dev =kzalloc (sizeof (*dev),GFP_KERNEL ); + struct gpio_sim_device *dev __free(kfree) = kzalloc(sizeof(*dev), +GFP_KERNEL );if (!dev)return ERR_PTR (-ENOMEM );
Signature · 1 line
2.39.2
Re: [PATCH v3 1/2] gpio: sim: fix an invalid __free() usage
In reply to Bartosz Golaszewskidrivers/gpio/gpio-sim.c gpio_sim_device_config_live_show(struct config_item *item, char *page) · line 733
Quoted 3 lines · Bartosz Golaszewski
drivers/gpio/gpio-sim.c +2
-0+ has_line_names = true; + max_offset =max (line->offset, max_offset);
I really don't understand why you kept this old broken logic.
I sent a much better version of this function that didn't need that pointless has_line_names thing or the 'max()' thing, by just making the code a lot simpler.
Whatever.
Quoted 4 lines
drivers/gpio/gpio-sim.c +4
-0+ line_names_size =gpio_sim_get_line_names_size (bank); +if (line_names_size) { + line_names =kcalloc (line_names_size,sizeof (*line_names), +GFP_KERNEL );
drivers/gpio/gpio-sim.c gpio_sim_make_bank_swnode(struct gpio_sim_bank *bank, · line 884
Quoted 10 lines
drivers/gpio/gpio-sim.c +5
-1+if (!line_names) +return ERR_PTR (-ENOMEM ); + +gpio_sim_set_line_names (bank, line_names); -if (line_names) properties[prop_idx++] =PROPERTY_ENTRY_STRING_ARRAY_LEN ("gpio-line-names" , line_names, line_names_size); + }
But I do like this reorganization.
Linus
Re: [PATCH v3 1/2] gpio: sim: fix an invalid __free() usage
Quoted 12 lines · Linus Torvalds
drivers/gpio/gpio-sim.c gpio_sim_device_config_live_show(struct config_item *item, char *page) · line 733
Quoted 3 lines · Bartosz Golaszewski
drivers/gpio/gpio-sim.c +2
-0+ has_line_names = true; + max_offset =max (line->offset, max_offset);I really don't understand why you kept this old broken logic.
I sent a much better version of this function that didn't need that pointless
has_line_namesthing or the 'max()' thing, by just making the code a lot simpler.
Right, it does what it's supposed to after all but IMO it's less clear, I had to take a second look now to get it. I was wondering if I'm simply too sleep deprived but no - it's because in your version the max_offset variable actually holds the value of (max_offset + 1) which makes the name untrue.
I don't want to bikeshed about it, let me know if my version is GoodEnough(R) or do you prefer another respin.
Bart
Quoted 20 lines
Whatever.
Quoted 4 lines
drivers/gpio/gpio-sim.c +4
-0+ line_names_size =gpio_sim_get_line_names_size (bank); +if (line_names_size) { + line_names =kcalloc (line_names_size,sizeof (*line_names), +GFP_KERNEL );drivers/gpio/gpio-sim.c gpio_sim_make_bank_swnode(struct gpio_sim_bank *bank, · line 884
Quoted 10 lines
drivers/gpio/gpio-sim.c +5
-1+if (!line_names) +return ERR_PTR (-ENOMEM ); + +gpio_sim_set_line_names (bank, line_names); -if (line_names) properties[prop_idx++] =PROPERTY_ENTRY_STRING_ARRAY_LEN ("gpio-line-names" , line_names, line_names_size); + }But I do like this reorganization.
Linus
Re: [PATCH v3 2/2] gpio: sim: initialize a managed pointer when declaring it
In reply to Bartosz GolaszewskiQuoted 6 lines · Bartosz Golaszewski
From: Bartosz Golaszewski <[email protected]>
Variables managed with
__free()should typically be initialized where they are declared so that the__free()callback is paired with its counterpart resource allocator. Fix the second instance of using__free()in gpio-sim to follow this pattern.
...
drivers/gpio/gpio-sim.c static const struct config_item_type gpio_sim_device_config_group_type = { · line 1489
Quoted 9 lines
{
drivers/gpio/gpio-sim.c +2
-2- struct gpio_sim_device *dev __free(kfree) = NULL;int id; - dev =kzalloc (sizeof (*dev),GFP_KERNEL ); + struct gpio_sim_device *dev __free(kfree) = kzalloc(sizeof(*dev), +GFP_KERNEL );if (!dev)return ERR_PTR (-ENOMEM );
Aside: Oh, this might be a downside of the __free() sugar, as we can theoretically end up with a code in the future like
struct bar *foo; ... struct baz *foo __free() = ... ...
and I am not sure how it goes to work. Or relaxed variant with
struct bar *foo;
...
{
struct baz *foo __free() = ...
...
}where we would have two variables with the same name, but different scope (this, perhaps, would work, but I assume compiler should warn about shadowed
name for the variable).
(Also what if in both cases bar == baz, i.e. same type?)
Signature · 2 lines
With Best Regards, Andy Shevchenko