Re: [PATCH v3] sunxi-irq: Fix Kconfig dependency on GENERIC_IRQ_CHIP

[Andre Przywara]

Hi Maxime,

On 02/05/16 07:48, Maxime Ripard wrote:
> Hi,
> 
> On Mon, Apr 25, 2016 at 02:04:52AM +0100, Andre Przywara wrote:
>> The Allwinner NMI irqchip driver requires GENERIC_IRQ_CHIP, but
>> we can't select it directly, because there is no specific Kconfig entry
>> for the driver. Compiling this NMI driver with certain arm64
>> configurations thus fails due to the missing dependency:
>>
>> drivers/built-in.o: In function `sunxi_sc_nmi_set_type':
>> drivers/irqchip/irq-sunxi-nmi.c:114: undefined reference to `irq_setup_alt_chip'
>> drivers/built-in.o: In function `irq_domain_add_linear':
>> include/linux/irqdomain.h:253: undefined reference to `irq_generic_chip_ops'
>> include/linux/irqdomain.h:253: undefined reference to `irq_generic_chip_ops'
>> drivers/built-in.o: In function `sunxi_sc_nmi_irq_init':
>> drivers/irqchip/irq-sunxi-nmi.c:146: undefined reference to `irq_alloc_domain_generic_chips'
>> drivers/irqchip/irq-sunxi-nmi.c:161: undefined reference to `irq_get_domain_generic_chip'
>> drivers/irqchip/irq-sunxi-nmi.c:170: undefined reference to `irq_gc_mask_clr_bit'
>> drivers/irqchip/irq-sunxi-nmi.c:171: undefined reference to `irq_gc_mask_set_bit'
>> drivers/irqchip/irq-sunxi-nmi.c:172: undefined reference to `irq_gc_ack_set_bit'
>> drivers/irqchip/irq-sunxi-nmi.c:170: undefined reference to `irq_gc_mask_clr_bit'
>>
>> Add separate Kconfig options for both Allwinner specific irqchip
>> drivers and select GENERIC_IRQ_CHIP for the NMI driver.
>> The older sun4i IRQ driver only gets selected when support for these
>> specific SoCs is compiled, while the NMI driver is selected for all
>> Allwinner SoCs (copying the current behaviour and covering future SoCs
>> automatically).
>>
>> Reported-by: Suzuki K Poulose <Suzuki.Poulose@xxxxxxx>
>> Signed-off-by: Andre Przywara <andre.przywara@xxxxxxx>
>> ---
>> Hi,
>>
>> this is a new approach to the problem that Suzuki tried to address
>> already.
>> It separates the sun4i IRQ controller driver and the NMI driver as
>> Maxime requested.
>> While there should be no difference for the NMI driver, the sun4i
>> driver now only gets selected for certain SoCs, which is admittedly
>> a bold call for -rc5. If people find this too risky, I can bring the
>> sun4i in line with the NMI driver (selected for all ARCH_SUNXI) for
>> this release and re-create the more selective dependency as a merge
>> window patch, so that it gets more testing.
>> Please let me know.
>>
>> Cheers,
>> Andre.
>>
>>  arch/arm/mach-sunxi/Kconfig | 3 ++-
>>  drivers/irqchip/Kconfig     | 8 ++++++++
>>  drivers/irqchip/Makefile    | 4 ++--
>>  3 files changed, 12 insertions(+), 3 deletions(-)
>>
>> diff --git a/arch/arm/mach-sunxi/Kconfig b/arch/arm/mach-sunxi/Kconfig
>> index c124d65..c092bc2 100644
>> --- a/arch/arm/mach-sunxi/Kconfig
>> +++ b/arch/arm/mach-sunxi/Kconfig
>> @@ -4,7 +4,6 @@ menuconfig ARCH_SUNXI
>>  	select ARCH_REQUIRE_GPIOLIB
>>  	select ARCH_HAS_RESET_CONTROLLER
>>  	select CLKSRC_MMIO
>> -	select GENERIC_IRQ_CHIP
>>  	select PINCTRL
>>  	select SUN4I_TIMER
>>  	select RESET_CONTROLLER
>> @@ -14,11 +13,13 @@ if ARCH_SUNXI
>>  config MACH_SUN4I
>>  	bool "Allwinner A10 (sun4i) SoCs support"
>>  	default ARCH_SUNXI
>> +	select SUN4I_IRQCHIP
>>  
>>  config MACH_SUN5I
>>  	bool "Allwinner A10s / A13 (sun5i) SoCs support"
>>  	default ARCH_SUNXI
>>  	select SUN5I_HSTIMER
>> +	select SUN4I_IRQCHIP
>>  
>>  config MACH_SUN6I
>>  	bool "Allwinner A31 (sun6i) SoCs support"
>> diff --git a/drivers/irqchip/Kconfig b/drivers/irqchip/Kconfig
>> index 3e12479..94f3f42 100644
>> --- a/drivers/irqchip/Kconfig
>> +++ b/drivers/irqchip/Kconfig
>> @@ -165,6 +165,14 @@ config ST_IRQCHIP
>>  	help
>>  	  Enables SysCfg Controlled IRQs on STi based platforms.
>>  
>> +config SUN4I_IRQCHIP
>> +	bool
>> +
>> +config SUNXI_NMI
>> +	bool "Allwinner SoC NMI controller"
>> +	default ARCH_SUNXI
>> +	select GENERIC_IRQ_CHIP
> 
> This one is only used on SUN6I, SUN7I and SUN8I. There's no need to
> enable it for all the SoCs.

But it is enabled for all SoCs at the moment - and as it is actually a
fix (-rc1 does not compile for certain arm64 configs) I'd really like to
not take chances here.
As mentioned I am happy to send a follow up patch to restrict it further
- given that it gets more testing, but at the moment I'd really like to
just fix the arm64 build without accidentally breaking existing arm(32)
boards.

Also as you mention sun8i here: AFAIK the H3, which I consider the
32-bit sibling of the A64, is an sun8i, so it will include the code, but
the A64 does not (atm) - which is a bit strange.
As I don't plan to have MACH_SUN<n>I defines for arm64 unless
desperately necessary, I'd really like to keep in on ARCH_SUNXI as it is
in the moment.

So how do we proceed from here?
Is this too late for 4.6 now? That would be a pity as this qualifies as
a fix. If not, I'd rather take this conservative patch for now and put a
more fine grained selection into -next (or 4.7-rc1).

Cheers,
Andre.