svn commit: r362338 - in head: share/man/man4 sys/conf sys/kern sys/netinet sys/netinet6 sys/netipsec sys/netpfil/pf

[John Baldwin]

On 6/22/20 3:07 PM, Mark Johnston wrote:
> On Mon, Jun 22, 2020 at 11:35:38AM -0700, John Baldwin wrote:
>> On 6/21/20 6:10 PM, Mark Johnston wrote:
>>> On Fri, Jun 19, 2020 at 08:33:35AM -0700, John Baldwin wrote:
>>>> On 6/18/20 12:32 PM, Mark Johnston wrote:
>>>>> Author: markj
>>>>> Date: Thu Jun 18 19:32:34 2020
>>>>> New Revision: 362338
>>>>> URL: https://svnweb.freebsd.org/changeset/base/362338
>>>>>
>>>>> Log:
>>>>>   Add the SCTP_SUPPORT kernel option.
>>>>>   
>>>>>   This is in preparation for enabling a loadable SCTP stack.  Analogous to
>>>>>   IPSEC/IPSEC_SUPPORT, the SCTP_SUPPORT kernel option must be configured
>>>>>   in order to support a loadable SCTP implementation.
>>>>>   
>>>>>   Discussed with:	tuexen
>>>>>   MFC after:	2 weeks
>>>>>   Sponsored by:	The FreeBSD Foundation
>>>>
>>>> Do you want to add similar handling to sys/conf/config.mk that we have
>>>> for IPsec?  Also, do we want to avoid building sctp.ko if it is in the
>>>> kernel like we do for ipsec.ko and/or only build it if the kernel contains
>>>> SCTP_SUPPORT?  (For ipsec.ko we had to do that as it wouldn't compile, not
>>>> sure if the same is true for sctp.ko)
>>>
>>> Sorry for the delay.
>>> I think we do indeed want similar handling in config.mk, I will work on
>>> it.  It is probably also reasonable to avoid compiling sctp.ko when
>>> SCTP_SUPPORT is not defined, though I can't see a reason that wouldn't
>>> work today since SCTP_SUPPORT is not used in any headers.
>>
>> Ok.  ipsec.ko mattered more when the build broke.  Whether or not we compile
>> "duplicate" modules for kernels is perhaps a larger question.  I think I
>> might favor that change, but it is a larger change that merits some thought.
>> In particular, you want good code coverage for things like LINT builds, so
>> maybe we really should still compile modules whenever possible.
> 
> I tend to assume that a buildkernel of GENERIC without any special flags
> will always build all modules (except those not available for the target
> platform of course), so I was a bit surprised to see that this isn't the
> case for ipsec.ko.  As Rodney pointed out it provides marginally better
> coverage against build breaks.  If you think we can restore the old
> behaviour for ipsec without too much work I think it'd be reasonable to
> change that and compile sctp.ko even when "options SCTP" is configured.
> I can't spot any similar cases in sys/modules/Makefile with a bit of
> skimming.

I don't think ipsec.ko is easily fixable when I looked at it.  I think it
is fine to leave sctp.ko building as part of GENERIC though.

-- 
John Baldwin