Re: [PATCH 1/3] cobalt: Prepare for new signature of __request_percpu_irq() in 6.19

On 16.02.26 14:44, Florian Bezdeka wrote:
> On Mon, 2026-02-16 at 12:43 +0100, Jan Kiszka wrote:
>> On 16.02.26 12:22, Florian Bezdeka wrote:
>>> On Mon, 2026-02-16 at 12:16 +0100, Jan Kiszka wrote:
>>>> On 16.02.26 12:15, Florian Bezdeka wrote:
>>>>> On Mon, 2026-02-16 at 12:11 +0100, Jan Kiszka wrote:
>>>>>> On 16.02.26 11:58, Florian Bezdeka wrote:
>>>>>>> Since 6.19 the signature of __request_percpu_irq() got one additional
>>>>>>> affinity parameter.
>>>>>>>
>>>>>>> A new Linux version specific wrapper cobalt_request_percpu_irq() is
>>>>>>> introduced to hide the signature change between different Linux
>>>>>>> versions.
>>>>>>>
>>>>>>> Signed-off-by: Florian Bezdeka <[email protected]>
>>>>>>> ---
>>>>>>>  include/cobalt/kernel/dovetail/pipeline/irq.h      | 20 ++++++++++++++++++++
>>>>>>>  include/cobalt/kernel/dovetail/pipeline/pipeline.h | 10 +++++-----
>>>>>>>  include/cobalt/kernel/dovetail/pipeline/sirq.h     |  8 +++-----
>>>>>>>  kernel/cobalt/dovetail/tick.c                      |  8 ++++----
>>>>>>>  4 files changed, 32 insertions(+), 14 deletions(-)
>>>>>>>
>>>>>>> diff --git a/include/cobalt/kernel/dovetail/pipeline/irq.h b/include/cobalt/kernel/dovetail/pipeline/irq.h
>>>>>>> index 55d9b8ff17cd08e5e8c09aec393a1e23736c1b76..ac780a4bb3223889ccf827b5b8cb279035a5dd7b 100644
>>>>>>> --- a/include/cobalt/kernel/dovetail/pipeline/irq.h
>>>>>>> +++ b/include/cobalt/kernel/dovetail/pipeline/irq.h
>>>>>>> @@ -5,6 +5,26 @@
>>>>>>>  #ifndef _COBALT_KERNEL_DOVETAIL_IRQ_H
>>>>>>>  #define _COBALT_KERNEL_DOVETAIL_IRQ_H
>>>>>>>  
>>>>>>> +#if LINUX_VERSION_CODE < KERNEL_VERSION(6, 19, 0)
>>>>>>> +static inline int cobalt_request_percpu_irq(unsigned int irq,
>>>>>>> +					    irq_handler_t handler,
>>>>>>> +					    unsigned long flags,
>>>>>>> +					    const char *devname,
>>>>>>> +					    void __percpu *dev_id)
>>>>>>> +{
>>>>>>> +	return __request_percpu_irq(irq, handler, flags, devname, dev_id);
>>>>>>> +}
>>>>>>> +#else
>>>>>>> +static inline int cobalt_request_percpu_irq(unsigned int irq,
>>>>>>> +					    irq_handler_t handler,
>>>>>>> +					    unsigned long flags,
>>>>>>> +					    const char *devname,
>>>>>>> +					    void __percpu *dev_id)
>>>>>>> +{
>>>>>>> +	return __request_percpu_irq(irq, handler, flags, devname, NULL, dev_id);
>>>>>>> +}
>>>>>>> +#endif
>>>>>>
>>>>>> Please use a define for request_percpu_irq on older kernels that uses
>>>>>> the new signature and resolves to the old one via __request_percpu_irq
>>>>>> as needed. 
>>>>>> Yes, also switch to request_percpu_irq directly because
>>>>>> __request_percpu_irq is going to be removed in 7.0.
>>>>>>
>>>>>
>>>>> request_percpu_irq() doesn't allow us to specify the flags. IRQF_OOB and
>>>>> IRQF_NO_THREAD would be needed.
>>>>
>>>> Then request_percpu_irq_affinity - whatever survives 7.0 already as well.
>>>>
>>>
>>> No flags as well. 
>>>
>>> I was not aware of __request_percpu_irq() being renamed in 7.0. Will try
>>> to find something else, but that doesn't look promising at the moment.
>>>
>>
>> Ouch, indeed, trouble ahead (or extra API needed from dovetail). If this
>> becomes a blocker moving forward with 6.19, just re-#define
>> __request_percpu_irq for now.
> 
> This is what this patch does, avoiding a strange function name override.
> 

This has has reason: When the kernel requiring the workaround has been
phased out, you do not have to touch all the call sites while dropping
the special cobal_* wrapping.

Jan

> Anyhow, as this way turned out to be not future prove I might have to
> provide an additional API in dovetail. Evl should have the same problem,
> so reuse seems likely.
> 

-- 
Siemens AG, Foundational Technologies
Linux Expert Center