Skip to contentSkip to conversation

power: bq27xxx_battery: Revert "Remove unneeded dependency in Kconfig"

PATCH In review Arnd Bergmann 2 participants 4 messages Nov 9, 2015 I2C subsystem host drivers

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 power: bq27xxx_battery: Revert "Remove unneeded dependency in Kconfig"

The dependency was clearly needed, without it it is possible to build the core i2c driver as a loadable module and the bq27xxx driver built-in, which results in link errors:

drivers/built-in.o: In function `bq27xxx_battery_i2c_read':
binder.c:(.text+0x360bf0): undefined reference to i2c_transfer' binder.c:(.text+0x360c10): undefined reference to i2c_transfer' drivers/built-in.o: In function `bq27xxx_battery_init':
binder.c:(.init.text+0xe668): undefined reference to i2c_register_driver' drivers/built-in.o: In function bq27xxx_battery_exit':
binder.c:(.exit.text+0x1a0c): undefined reference to `i2c_del_driver'

Fixes 6bd03ce3c12a ("power: bq27xxx_battery: Remove unneeded dependency in Kconfig")

---
The bug was originally found and fixed by Xiong Zhou, but Andrew Davis broke it again by reverting the fix.

I found it today on my ARM randconfig builds.

drivers/power/Kconfig +1 -0
index 237d7aa73e8c..9f53fb74ae6f 100644
@@ -159,6 +159,7 @@ config BATTERY_SBS
 
 config BATTERY_BQ27XXX
 	tristate "BQ27xxx battery driver"
+	depends on I2C || I2C=n
 	help
 	  Say Y here to enable support for batteries with BQ27xxx (I2C/HDQ) chips.
 
#1 lore raw
Quoted 26 lines · Arnd Bergmann

The dependency was clearly needed, without it it is possible to build the core i2c driver as a loadable module and the bq27xxx driver built-in, which results in link errors:

drivers/built-in.o: In function `bq27xxx_battery_i2c_read':
binder.c:(.text+0x360bf0): undefined reference to i2c_transfer' binder.c:(.text+0x360c10): undefined reference to i2c_transfer' drivers/built-in.o: In function `bq27xxx_battery_init':
binder.c:(.init.text+0xe668): undefined reference to i2c_register_driver' drivers/built-in.o: In function bq27xxx_battery_exit':
binder.c:(.exit.text+0x1a0c): undefined reference to `i2c_del_driver'

Fixes 6bd03ce3c12a ("power: bq27xxx_battery: Remove unneeded dependency in Kconfig")

---
The bug was originally found and fixed by Xiong Zhou, but Andrew Davis broke it again by reverting the fix.

I found it today on my ARM randconfig builds.

drivers/power/Kconfig +0 -0
index 237d7aa73e8c..9f53fb74ae6f 100644

drivers/power/Kconfig config BATTERY_SBS · line 165

Quoted 7 lines
drivers/power/Kconfig +1 -0
@@ -159,6 +159,7 @@ config BATTERY_SBS
 
  config BATTERY_BQ27XXX
  	tristate "BQ27xxx battery driver"
+	depends on I2C || I2C=n
  	help
  	  Say Y here to enable support for batteries with BQ27xxx (I2C/HDQ) chips.
 

Nothing enabled by BATTERY_BQ27XXX depends on I2C, this workaround is not correct as it prevents BATTERY_BQ27XXX from being built-in when I2C is a module, there is no reason for this limitation.

The undefined references are caused by BATTERY_BQ27XXX being built-in AND its I2C functionality being enabled (BATTERY_BQ27XXX_I2C) while I2C is a module. Reorganizing this driver is being discussed anyway, but in the meantime a more correct fix would be along the lines of:

drivers/power/Kconfig +1 -0
index 6de6ec2..d1d32f9 100644
@@ -167,6 +167,7 @@ config BATTERY_BQ27XXX_I2C
         bool "BQ27xxx I2C support"
         depends on BATTERY_BQ27XXX
         depends on I2C
+       depends on !(I2C=m && BATTERY_BQ27XXX=y)
         default y
         help
           Say Y here to enable support for batteries with BQ27xxx (I2C) chips.
#2 lore raw
Quoted 15 lines · On Monday 09 November 2015 10:56:13 Andrew F. Davis wrote:

On 11/09/2015 07:50 AM, Arnd Bergmann wrote:
Nothing enabled by BATTERY_BQ27XXX depends on I2C, this workaround is not correct as it prevents BATTERY_BQ27XXX from being built-in when I2C is a module, there is no reason for this limitation.

The undefined references are caused by BATTERY_BQ27XXX being built-in AND its I2C functionality being enabled (BATTERY_BQ27XXX_I2C) while I2C is a module. Reorganizing this driver is being discussed anyway, but in the meantime a more correct fix would be along the lines of:

drivers/power/Kconfig +0 -0
index 6de6ec2..d1d32f9 100644

drivers/power/Kconfig config BATTERY_BQ27XXX_I2C · line 173

Quoted 7 lines
drivers/power/Kconfig +1 -0
@@ -167,6 +167,7 @@ config BATTERY_BQ27XXX_I2C
         bool "BQ27xxx I2C support"
         depends on BATTERY_BQ27XXX
         depends on I2C
+       depends on !(I2C=m && BATTERY_BQ27XXX=y)
         default y
         help
           Say Y here to enable support for batteries with BQ27xxx (I2C) chips.

That works too, there is just very little difference in the end here, and it's easier to revert an patch that only introduces a regression than to do a different hack, especially if it's going to be reworked soon anyway.

Do you want to submit the above as a fixup to your other patch or should we just do the revert? It would be good to get one of the two into -rc1.

Arnd

#3 lore raw
Quoted 29 lines · Arnd Bergmann
Quoted 15 lines · On Monday 09 November 2015 10:56:13 Andrew F. Davis wrote:

On 11/09/2015 07:50 AM, Arnd Bergmann wrote:
Nothing enabled by BATTERY_BQ27XXX depends on I2C, this workaround is not correct as it prevents BATTERY_BQ27XXX from being built-in when I2C is a module, there is no reason for this limitation.

The undefined references are caused by BATTERY_BQ27XXX being built-in AND its I2C functionality being enabled (BATTERY_BQ27XXX_I2C) while I2C is a module. Reorganizing this driver is being discussed anyway, but in the meantime a more correct fix would be along the lines of:

drivers/power/Kconfig +0 -0
index 6de6ec2..d1d32f9 100644

drivers/power/Kconfig config BATTERY_BQ27XXX_I2C · line 173

Quoted 7 lines
drivers/power/Kconfig +1 -0
@@ -167,6 +167,7 @@ config BATTERY_BQ27XXX_I2C
          bool "BQ27xxx I2C support"
          depends on BATTERY_BQ27XXX
          depends on I2C
+       depends on !(I2C=m && BATTERY_BQ27XXX=y)
          default y
          help
            Say Y here to enable support for batteries with BQ27xxx (I2C) chips.

That works too, there is just very little difference in the end here, and it's easier to revert an patch that only introduces a regression than to do a different hack, especially if it's going to be reworked soon anyway.

Quoted 4 lines

Do you want to submit the above as a fixup to your other patch or should we just do the revert? It would be good to get one of the two into -rc1.

I can just submit this fix, it would probably be better for the short term until the rework is done.

Andrew