Re: [Ksummit-discuss] [CORE TOPIC] Reviewing new API/ABI

[Jiri Kosina <jkosina@suse.cz>]

On Thu, 8 May 2014, Li Zefan wrote:

> >> diff --git a/MAINTAINERS b/MAINTAINERS
> >> index 7578deb..52a282b 100644
> >> --- a/MAINTAINERS
> >> +++ b/MAINTAINERS
> >> @@ -210,6 +210,13 @@ S:	Supported
> >>  F:	Documentation/scsi/aacraid.txt
> >>  F:	drivers/scsi/aacraid/
> >>  
> >> +ABI/API
> >> +L:	linux-api@vger.kernel.org
> >> +F:	Documentation/ABI/
> >> +F:	include/linux/syscalls.h
> >> +F:	include/uapi/
> >> +F:	kernel/sys_ni.c
> >> +
> > 
> > It'd be nice to have also sysfs changes covered as well if this is not 
> > just about API, but also ABI; I am not sure whether this could be covered 
> > by MAINTAINERS pattern, but at least mentioning this list in 
> > Documentation/ABI/README seems appropriate.
> 
> If sysfs is in, 

Well, sysfs definitely is considered kernel ABI.

> what about cgroupfs? Which has been under dramatic changes.

This is of course interesting (and important) question per se -- the 
amount of interfaces between userspace and kernelspace is growing at a 
rapid pace (especially a lot of crap gets added into debugfs I guess), and 
it's not always clearly defined what is considered proper ABI and thus 
should be maintained stable.

Is "if we ever get a report about userspace regression because of kernel 
interface change" enough well-defined criteria? What if there is userspace 
depending on the contents of the kernel ringbuffer? Etc etc ...

-- 
Jiri Kosina
SUSE Labs