clk: rockchip: cleanup errors in (module-)driver handling
Subsystems (2)
Heiko Stuebner
maintainercover clk: rockchip: cleanup errors in (module-)driver handling
Recent conversions or inclusions of rk3399 and rk3568 as platform-drivers and subsequently allowing them to be built as modules introduced some problems.
These two patches try to correct them.
Heiko Stuebner (2):
Revert "clk: rockchip: use module_platform_driver_probe"
clk: rockchip: drop module parts from rk3399 and rk3568 drivers
drivers/clk/rockchip/Kconfig | 4 ++-- drivers/clk/rockchip/clk-rk3399.c | 6 +----- drivers/clk/rockchip/clk-rk3568.c | 6 +----- 3 files changed, 4 insertions(+), 12 deletions(-)
Signature · 1 line
2.30.2
Heiko Stuebner
maintainerPATCH 1/2 Revert "clk: rockchip: use module_platform_driver_probe"
This reverts commit 1da80da028fe5accb866c0d6899a292ed86bef45.
Reading recent discussions [0] [1], I realized this change introduces a number of problems:
- only converting to module_platform_driver creates the issue with the existing __init and __initdata attributes.
When the driver would've been built as a module, all the missing clock-definitions (all are initdata) should've turned up as error in testing suggesting that the change wasn't at all
- a clock driver is a very core component of soc bringup and making this able to be built as a module solely for enabling the soc vendor to add out of tree changes for Android implementations is not in our interest and also everything except a ramdisk won't probe without a clock controller.
This is especially true when the changes aren't really tested and are merely added to move the mainline driver "out of the way".
[0] https://lwn.net/Articles/872209/
[1] https://lore.kernel.org/all/[email protected]/
--- drivers/clk/rockchip/clk-rk3399.c | 2 +- drivers/clk/rockchip/clk-rk3568.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
drivers/clk/rockchip/clk-rk3399.c +1 -1
@@ -1656,7 +1656,7 @@ static struct platform_driver clk_rk3399_driver = { .suppress_bind_attrs = true, }, }; -module_platform_driver_probe (clk_rk3399_driver, clk_rk3399_probe); +builtin_platform_driver_probe (clk_rk3399_driver, clk_rk3399_probe);MODULE_DESCRIPTION ("Rockchip RK3399 Clock Driver" );MODULE_LICENSE ("GPL" );
drivers/clk/rockchip/clk-rk3568.c +1 -1
@@ -1719,7 +1719,7 @@ static struct platform_driver clk_rk3568_driver = { .suppress_bind_attrs = true, }, }; -module_platform_driver_probe (clk_rk3568_driver, clk_rk3568_probe); +builtin_platform_driver_probe (clk_rk3568_driver, clk_rk3568_probe);MODULE_DESCRIPTION ("Rockchip RK3568 Clock Driver" );MODULE_LICENSE ("GPL" );
Signature · 1 line
2.30.2
Re: [PATCH 1/2] Revert "clk: rockchip: use module_platform_driver_probe"
Quoting Heiko Stuebner (2021-10-27 06:26:15)
Quoted 18 lines
This reverts commit
1da80da028fe5accb866c0d6899a292ed86bef45.Reading recent discussions [0] [1], I realized this change introduces a number of problems:
- only converting to
module_platform_drivercreates the issue with the existing__initand__initdataattributes.When the driver would've been built as a module, all the missing clock-definitions (all are initdata) should've turned up as error in testing suggesting that the change wasn't at all
- a clock driver is a very core component of soc bringup and making this able to be built as a module solely for enabling the soc vendor to add out of tree changes for Android implementations is not in our interest and also everything except a ramdisk won't probe without a clock controller.
Quoted 8 lines
This is especially true when the changes aren't really tested and are merely added to move the mainline driver "out of the way".
[0] https://lwn.net/Articles/872209/
[1] https://lore.kernel.org/all/[email protected]/Signed-off-by Heiko Stuebner <[email protected]>---
Heiko Stuebner
maintainerPATCH 2/2 clk: rockchip: drop module parts from rk3399 and rk3568 drivers
In reply to Heiko StuebnerBoth of these drivers were converted to real drivers and got a tristate build option. But them being builtin_platform_drivers, they only ever should be build-in - as the name suggests.
So adapt the Kconfig symbol and drop the MODULE_* parts from the drivers.
--- drivers/clk/rockchip/Kconfig | 4 ++-- drivers/clk/rockchip/clk-rk3399.c | 4 ---- drivers/clk/rockchip/clk-rk3568.c | 4 ---- 3 files changed, 2 insertions(+), 10 deletions(-)
drivers/clk/rockchip/Kconfig +2 -2
@@ -80,14 +80,14 @@ config CLK_RK3368 Build the driver for RK3368 Clock Driver. config CLK_RK3399 - tristate "Rockchip RK3399 clock controller support" + bool "Rockchip RK3399 clock controller support" depends on ARM64 || COMPILE_TEST default y help Build the driver for RK3399 Clock Driver. config CLK_RK3568 - tristate "Rockchip RK3568 clock controller support" + bool "Rockchip RK3568 clock controller support" depends on ARM64 || COMPILE_TEST default y help
drivers/clk/rockchip/clk-rk3399.c +0 -4
@@ -1630,7 +1630,6 @@ static const struct of_device_id clk_rk3399_match_table[] = { }, { } }; -MODULE_DEVICE_TABLE (of, clk_rk3399_match_table);static int __init clk_rk3399_probe (struct platform_device *pdev) {
@@ -1657,6 +1656,3 @@ static struct platform_driver clk_rk3399_driver = { }, };builtin_platform_driver_probe (clk_rk3399_driver, clk_rk3399_probe); - -MODULE_DESCRIPTION ("Rockchip RK3399 Clock Driver" ); -MODULE_LICENSE ("GPL" );
drivers/clk/rockchip/clk-rk3568.c +0 -4
@@ -1693,7 +1693,6 @@ static const struct of_device_id clk_rk3568_match_table[] = { }, { } }; -MODULE_DEVICE_TABLE (of, clk_rk3568_match_table);static int __init clk_rk3568_probe (struct platform_device *pdev) {
@@ -1720,6 +1719,3 @@ static struct platform_driver clk_rk3568_driver = { }, };builtin_platform_driver_probe (clk_rk3568_driver, clk_rk3568_probe); - -MODULE_DESCRIPTION ("Rockchip RK3568 Clock Driver" ); -MODULE_LICENSE ("GPL" );
Signature · 1 line
2.30.2
Re: [PATCH 2/2] clk: rockchip: drop module parts from rk3399 and rk3568 drivers
Quoting Heiko Stuebner (2021-10-27 06:26:16)
Quoted 8 lines
Both of these drivers were converted to real drivers and got a tristate build option. But them being
builtin_platform_drivers, they only ever should be build-in - as the name suggests.So adapt the
Kconfigsymbol and drop the MODULE_* parts from the drivers.Signed-off-by Heiko Stuebner <[email protected]>---
Stephen Boyd
maintainerQuoting Heiko Stuebner (2021-10-27 06:26:14)
Quoted 5 lines
Recent conversions or inclusions of rk3399 and rk3568 as platform-drivers and subsequently allowing them to be built as modules introduced some problems.
These two patches try to correct them.
By removing modular support? Ok.
Heiko Stübner
maintainerAm Mittwoch, 27. Oktober 2021, 22:27:46 CEST schrieb Stephen Boyd:
Quoted 8 lines
Quoting Heiko Stuebner (2021-10-27 06:26:14)
Quoted 5 lines
Recent conversions or inclusions of rk3399 and rk3568 as platform-drivers and subsequently allowing them to be built as modules introduced some problems.
These two patches try to correct them.
By removing modular support? Ok.
It looked like the easiest way to go for now.
With all the clock-definitions as init-data still in both drivers, I don't think anybody ever tested running this as a real module.
We could of course also remove all the __init + __initdata attributes, but still wouldn't know if it actually runs as a module, without someone building a real test environment for it.
Heiko
Stephen Boyd
maintainerQuoting Heiko Stübner (2021-10-27 16:20:17)
Quoted 10 lines
Am Mittwoch, 27. Oktober 2021, 22:27:46 CEST schrieb Stephen Boyd:
Quoted 8 lines
Quoting Heiko Stuebner (2021-10-27 06:26:14)
Quoted 5 lines
Recent conversions or inclusions of rk3399 and rk3568 as platform-drivers and subsequently allowing them to be built as modules introduced some problems.
These two patches try to correct them.
By removing modular support? Ok.
Quoted 10 lines
It looked like the easiest way to go for now.
With all the clock-definitions as init-data still in both drivers, I don't think anybody ever tested running this as a real module.
We could of course also remove all the
__init+__initdataattributes, but still wouldn't know if it actually runs as a module, without someone building a real test environment for it.
Sure. Do I need to pick these up directly?
Heiko Stübner
maintainerAm Freitag, 29. Oktober 2021, 08:43:05 CEST schrieb Stephen Boyd:
Quoted 16 lines
Quoting Heiko Stübner (2021-10-27 16:20:17)
Quoted 15 lines
Am Mittwoch, 27. Oktober 2021, 22:27:46 CEST schrieb Stephen Boyd:
Quoted 8 lines
Quoting Heiko Stuebner (2021-10-27 06:26:14)
Quoted 5 lines
Recent conversions or inclusions of rk3399 and rk3568 as platform-drivers and subsequently allowing them to be built as modules introduced some problems.
These two patches try to correct them.
By removing modular support? Ok.
It looked like the easiest way to go for now.
With all the clock-definitions as init-data still in both drivers, I don't think anybody ever tested running this as a real module.
Quoted 7 lines
Quoted 5 lines
We could of course also remove all the
__init+__initdataattributes, but still wouldn't know if it actually runs as a module, without someone building a real test environment for it.Sure. Do I need to pick these up directly?
I don't have anything more for 5.16 and the 5.15 release should happen on sunday, so yeah just pick them if you like - no need for doing a separate pull request.
Thanks
Heiko
Stephen Boyd
maintainerQuoting Heiko Stübner (2021-10-29 05:37:06)
Quoted 23 lines
Am Freitag, 29. Oktober 2021, 08:43:05 CEST schrieb Stephen Boyd:
Quoted 22 lines
Quoting Heiko St�bner (2021-10-27 16:20:17)
Quoted 20 lines
Am Mittwoch, 27. Oktober 2021, 22:27:46 CEST schrieb Stephen Boyd:
Quoted 8 lines
Quoting Heiko Stuebner (2021-10-27 06:26:14)
Quoted 5 lines
Recent conversions or inclusions of rk3399 and rk3568 as platform-drivers and subsequently allowing them to be built as modules introduced some problems.
These two patches try to correct them.
By removing modular support? Ok.
It looked like the easiest way to go for now.
With all the clock-definitions as init-data still in both drivers, I don't think anybody ever tested running this as a real module.
We could of course also remove all the
__init+__initdataattributes, but still wouldn't know if it actually runs as a module, without someone building a real test environment for it.
Quoted 6 lines
Quoted 1 line
Sure. Do I need to pick these up directly?
I don't have anything more for 5.16 and the 5.15 release should happen on sunday, so yeah just pick them if you like - no need for doing a separate pull request.
Ok I picked them up and I'll send them off next week.