Re: [PATCH 1/3] cobalt uapi: Introducing new feature flag for settime64 availability

[Jan Kiszka <jan.kiszka@siemens.com>]

On 14.11.20 07:57, song via Xenomai wrote:
> 
> 
> On 2020/11/13 下午8:00, florian.bezdeka@siemens.com wrote:
>> From: Florian Bezdeka <florian.bezdeka@siemens.com>
>>
>> Adding a new feature flag to allow the library asking for settime64
>> support. That will allow the library to use the new system call if it
>> is available / supported by the kernel.
>>
>> Signed-off-by: Florian Bezdeka <florian.bezdeka@siemens.com>
>> ---
>>   include/cobalt/uapi/asm-generic/features.h | 6 +++++-
>>   1 file changed, 5 insertions(+), 1 deletion(-)
>>
>> diff --git a/include/cobalt/uapi/asm-generic/features.h
>> b/include/cobalt/uapi/asm-generic/features.h
>> index 8a4927c49..93a96670a 100644
>> --- a/include/cobalt/uapi/asm-generic/features.h
>> +++ b/include/cobalt/uapi/asm-generic/features.h
>> @@ -51,6 +51,7 @@ struct cobalt_featinfo {
>>   #define __xn_feat_nofastsynch 0x10000000
>>   #define __xn_feat_control     0x08000000
>>   #define __xn_feat_prioceiling 0x04000000
>> +#define __xn_feat_settime64   0x02000000
>>     #ifdef CONFIG_SMP
>>   #define __xn_feat_smp_mask __xn_feat_smp
>> @@ -70,7 +71,8 @@ struct cobalt_featinfo {
>>   #define __xn_feat_generic_mask            \
>>       (__xn_feat_smp_mask        |    \
>>        __xn_feat_fastsynch_mask     |    \
>> -     __xn_feat_prioceiling)
>> +     __xn_feat_prioceiling        |    \
>> +     __xn_feat_settime64 )
>>     /*
>>    * List of features both sides have to agree on: If userland supports
>> @@ -101,6 +103,8 @@ const char *get_generic_feature_label(unsigned int
>> feature)
>>           return "control";
>>       case __xn_feat_prioceiling:
>>           return "prioceiling";
>> +    case __xn_feat_settime64:
>> +        return "settime64";
>>       default:
>>           return 0;
>>       }
>>
> 
> If i understand it correctly, we negotiate xn_feature instead of abi
> revision? In lib/cobalt/clock.c?
> 

Yes. New userspace will mandate the feature, old will not require it.
Kernel can handle both cases, passively.

> Another question is settime64 is the first step of y2038, clock_gettime
> and others will come up, shall we have xn_feat_time64 for all of them?

Yes, good point.

Jan
-- 
Siemens AG, T RDA IOT
Corporate Competence Center Embedded Linux