Skip to contentSkip to conversation

mfd: syscon: Removed support for unloading

PATCH In review Alexander Shiyan 4 participants 10 messages Feb 23, 2013 – Feb 25, 2013

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH v5 1/3 mfd: syscon: Removed support for unloading

The driver can be used in various subsystems and therefore should not be unloaded when it is defined in the kernel configuration, so remove support for unloading it.

---
 drivers/mfd/syscon.c | 18 ------------------
 1 file changed, 18 deletions(-)
drivers/mfd/syscon.c +0 -18
index 61aea63..55d7915 100644
@@ -139,17 +139,6 @@ static int syscon_probe(struct platform_device *pdev)
 	return 0;
 }
 
-static int syscon_remove(struct platform_device *pdev)
-{
-	struct syscon *syscon;
-
-	syscon = platform_get_drvdata(pdev);
-	iounmap(syscon->base);
-	platform_set_drvdata(pdev, NULL);
-
-	return 0;
-}
-
 static struct platform_driver syscon_driver = {
 	.driver = {
 		.name = "syscon",
@@ -157,7 +146,6 @@ static struct platform_driver syscon_driver = {
 		.of_match_table = of_syscon_match,
 	},
 	.probe		= syscon_probe,
-	.remove		= syscon_remove,
 };
 
 static int __init syscon_init(void)
@@ -166,12 +154,6 @@ static int __init syscon_init(void)
 }
 postcore_initcall(syscon_init);
 
-static void __exit syscon_exit(void)
-{
-	platform_driver_unregister(&syscon_driver);
-}
-module_exit(syscon_exit);
-
 MODULE_AUTHOR("Dong Aisheng <[email protected]>");
 MODULE_DESCRIPTION("System Control driver");
 MODULE_LICENSE("GPL v2");
Signature · 1 line
1.7.12.4
#1 lore raw

PATCH v5 2/3 mfd: syscon: Removed unneeded field "dev" from private driver structure

---
 drivers/mfd/syscon.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)
drivers/mfd/syscon.c +1 -4
index 55d7915..2c59ce6 100644
@@ -25,17 +25,15 @@
 static struct platform_driver syscon_driver;
 
 struct syscon {
-	struct device *dev;
 	void __iomem *base;
 	struct regmap *regmap;
 };
 
 static int syscon_match(struct device *dev, void *data)
 {
-	struct syscon *syscon = dev_get_drvdata(dev);
 	struct device_node *dn = data;
 
-	return (syscon->dev->of_node == dn) ? 1 : 0;
+	return (dev->of_node == dn) ? 1 : 0;
 }
 
 struct regmap *syscon_node_to_regmap(struct device_node *np)
@@ -130,7 +128,6 @@ static int syscon_probe(struct platform_device *pdev)
 		return PTR_ERR(syscon->regmap);
 	}
 
-	syscon->dev = dev;
 	platform_set_drvdata(pdev, syscon);
 
 	dev_info(dev, "syscon regmap start 0x%x end 0x%x registered\n",
Signature · 1 line
1.7.12.4
#2 lore raw
In reply to Alexander Shiyan
Quoted 3 lines · Alexander Shiyan

The driver can be used in various subsystems and therefore should not be unloaded when it is defined in the kernel configuration, so remove support for unloading it.

Why not fix the clients to module_get() at the appropriate times; then you could still allow unloading, couldn't you?

#3 lore raw
Quoted 7 lines · Alexander Shiyan
Quoted 3 lines

The driver can be used in various subsystems and therefore should not be unloaded when it is defined in the kernel configuration, so remove support for unloading it.

Why not fix the clients to module_get() at the appropriate times; then you could still allow unloading, couldn't you?

I has explain this before. Driver defined as "bool" and loaded via postcore_initcall. Once loaded it should not be unloaded, in other case it have not way to be loaded back. I am correctly understand your question?
Thanks.

---
���{.n�+�������+%�����ݶ��w��{.n�+����{��G�����{ay�ʇڙ�,j��f���h��������z_�(�階�ݢj"���m������G������?����&���~��iO��z��v�^�m���� ������?�I�

#4 lore raw
Quoted 9 lines · Alexander Shiyan
Quoted 7 lines · Alexander Shiyan
Quoted 3 lines

The driver can be used in various subsystems and therefore should not be unloaded when it is defined in the kernel configuration, so remove support for unloading it.

Why not fix the clients to module_get() at the appropriate times; then you could still allow unloading, couldn't you?

I has explain this before.

If multiple people have asked this, perhaps it'd be a good idea to include the answer in the commit description.

Quoted 1 line

Driver defined as "bool" and loaded via postcore_initcall.

Being defined as a "bool" sounds like a reasonable reason that no remove() is required.

Being loaded via postcore_initcall() (a) should have absolutely no influence over whether a remove() is required (an actual module could just as well include a postcore_initcall function, which would get executed at module load time), and (b) is wrong; initcall ordering shouldn't be used to influence driver probe ordering.

#5 lore raw
Quoted 11 lines · Stephen Warren
Quoted 9 lines · Alexander Shiyan
Quoted 7 lines · Alexander Shiyan
Quoted 3 lines

The driver can be used in various subsystems and therefore should not be unloaded when it is defined in the kernel configuration, so remove support for unloading it.

Why not fix the clients to module_get() at the appropriate times; then you could still allow unloading, couldn't you?

I has explain this before.

Quoted 7 lines

If multiple people have asked this, perhaps it'd be a good idea to include the answer in the commit description.

Quoted 1 line

Driver defined as "bool" and loaded via postcore_initcall.

Being defined as a "bool" sounds like a reasonable reason that no remove() is required.

No, it is not - I can still unbind the device from driver via sysfs. Now what will happen is resources are leaked and won't be available when trying to bin (via sysfs) again.

This patch is broken.

Thanks.

Signature · 1 line
Dmitry
#6 lore raw
Quoted 17 lines · Stephen Warren
Quoted 16 lines · Alexander Shiyan
Quoted 9 lines
Quoted 7 lines · Alexander Shiyan
Quoted 3 lines

The driver can be used in various subsystems and therefore should not be unloaded when it is defined in the kernel configuration, so remove support for unloading it.

Why not fix the clients to module_get() at the appropriate times; then you could still allow unloading, couldn't you?

I has explain this before.

If multiple people have asked this, perhaps it'd be a good idea to include the answer in the commit description.

Quoted 1 line

Driver defined as "bool" and loaded via postcore_initcall.

Quoted 8 lines
Quoted 2 lines

Being defined as a "bool" sounds like a reasonable reason that no remove() is required.

No, it is not - I can still unbind the device from driver via sysfs. Now what will happen is resources are leaked and won't be available when trying to bin (via sysfs) again.

This patch is broken.

I will try to resolve the dispute.
Initially, the first part of the patch (this) was an attempt to prevent the unloading (unregister) of the driver (not the device), ie remove module_exit call. The patch also removes the code to remove the device. Inclusion in this patch of the code was a mistake. Because of this, people are confused about the terms, including me.
Code, relating to the removal of the device should be moved to the third patch. Since all moved to using the managed resources, it should not have any questions.

About uloading (unregister) driver: As far i understand "__exit" section is completely discarded when kernel compiled without module support. But where is this call (module_exit) is placing in case "bool" driver and kernel is compiled with module support? Fixme please.

Thanks.

---
���{.n�+�������+%�����ݶ��w��{.n�+����{��G�����{ay�ʇڙ�,j��f���h��������z_�(�階�ݢj"���m������G������?����&���~��iO��z��v�^�m���� ������?�I�

#7 lore raw
Quoted 35 lines · Alexander Shiyan
Quoted 17 lines · Stephen Warren
Quoted 16 lines · Alexander Shiyan
Quoted 9 lines
Quoted 7 lines · Alexander Shiyan
Quoted 3 lines

The driver can be used in various subsystems and therefore should not be unloaded when it is defined in the kernel configuration, so remove support for unloading it.

Why not fix the clients to module_get() at the appropriate times; then you could still allow unloading, couldn't you?

I has explain this before.

If multiple people have asked this, perhaps it'd be a good idea to include the answer in the commit description.

Quoted 1 line

Driver defined as "bool" and loaded via postcore_initcall.

Quoted 8 lines
Quoted 2 lines

Being defined as a "bool" sounds like a reasonable reason that no remove() is required.

No, it is not - I can still unbind the device from driver via sysfs. Now what will happen is resources are leaked and won't be available when trying to bin (via sysfs) again.

This patch is broken.

I will try to resolve the dispute.
Initially, the first part of the patch (this) was an attempt to prevent the unloading (unregister) of the driver (not the device), ie remove module_exit call. The patch also removes the code to remove the device. Inclusion in this patch of the code was a mistake. Because of this, people are confused about the terms, including me.
Code, relating to the removal of the device should be moved to the third patch. Since all moved to using the managed resources, it should not have any questions.

Quoted 4 lines

About uloading (unregister) driver: As far i understand "__exit" section is completely discarded when kernel compiled without module support. But where is this call (module_exit) is placing in case "bool" driver and kernel is compiled with module support? Fixme please.

In both cases __exit code goes to .exit.text section which is either discarded (code is compiled into the kernel) or kept (code is built as a module).

Still, because the device can forcibly be unbound from the driver, all these changes are pretty useless for the stated purpose. All in all, it seems to me that this code should not be a driver at all, but rather rolled into regmap infrastructure and be part of it's initialization process.

Thanks.

Signature · 1 line
Dmitry
#8 lore raw

PATCH v5 3/3 mfd: syscon: Add non-DT support

In reply to Alexander Shiyan

This patch allow using syscon driver from the platform data, i.e. possibility using driver on systems without oftree support. For search syscon device from the client drivers, "syscon_regmap_lookup_by_pdevname" function was added.

---
 drivers/mfd/Kconfig        |  1 -
 drivers/mfd/syscon.c       | 67 ++++++++++++++++++++++++++++++++--------------
 include/linux/mfd/syscon.h |  1 +
 3 files changed, 48 insertions(+), 21 deletions(-)
drivers/mfd/Kconfig +0 -1
index 671f5b1..8fdd87e 100644
@@ -1070,7 +1070,6 @@ config MFD_STA2X11
 
 config MFD_SYSCON
 	bool "System Controller Register R/W Based on Regmap"
-	depends on OF
 	select REGMAP_MMIO
 	help
 	  Select this option to enable accessing system control registers
drivers/mfd/syscon.c +47 -20

115 lines not on this page · raw message

include/linux/mfd/syscon.h +1 -0
index 6aeb6b8..5c9ee6e 100644
@@ -17,6 +17,7 @@
 
 extern struct regmap *syscon_node_to_regmap(struct device_node *np);
 extern struct regmap *syscon_regmap_lookup_by_compatible(const char *s);
+extern struct regmap *syscon_regmap_lookup_by_pdevname(const char *s);
 extern struct regmap *syscon_regmap_lookup_by_phandle(
 					struct device_node *np,
 					const char *property);
Signature · 1 line
1.7.12.4