Fix LINT-NOINET kernels
- Generate opt_inet.h appropriately according to MK_INET_SUPPORT
- Only add in_gif.c to SRCS if MK_INET_SUPPORT != no
MFC after: 1 week
Sponsored by: EMCC / Isilon Storage Division
Details
Details
- Reviewers
ae
I built the kernel module like so and ran..
ifconfig gif0 create ifconfig gif0 destroy
.. without error:
INET | INET6 no | no yes | no no | yes yes | yes
make tinderbox passed (finally) on universe10a.freebsd.org:
--------------------------------------------------------------
>>> make universe completed on Tue Jan 12 05:46:57 UTC 2016
(started Tue Jan 12 00:48:41 UTC 2016)
--------------------------------------------------------------Diff Detail
Diff Detail
- Repository
- rS FreeBSD src repository - subversion
- Lint
Lint Passed - Unit
No Test Coverage - Build Status
Buildable 2048 Build 2056: arc lint + arc unit
Event Timeline
| sys/modules/if_gif/Makefile | ||
|---|---|---|
| 16 | opt_inet.h can contain several options, e.g. TCP_OFFLOAD, TCP_SIGNATURE. So, in general this expression can give wrong result. | |
| sys/modules/if_gif/Makefile | ||
|---|---|---|
| 16 | I agree in principle, but disagree for the following reasons:
$ grep -r TCP_ ../../*/*_gif.c || echo not found not found
I'm all for fixing this up, but I would rather invest time in backporting config.mk as it makes a lot more sense than dealing with piecemeal backports of opt_net.h. | |
| 20–22 | I'll remove lines 20 and 22 to fold the expressions into one expression | |
Comment Actions
This is how it works in my understanding. You have builded LINT-NOINET kernel. opt_inet.h in KERNBUILDDIR will contain TCP_SIGNATURE option and when modules will be build, this expression will give wrong result. I.e. it will try build kernel module with INET.
Comment Actions
In this case though, opt_inet.h (if it doesn't already exist in /usr/obj/sys/modules/if_gif/) will only contain #define INET 1, or will be empty.
Comment Actions
Just tried:
# pwd /usr/obj/home/devel/freebsd/base/stable/10/sys/LINT-NOINET/modules/home/devel/freebsd/base/stable/10/sys/modules/if_gif # ll total 30 -rw-r--r-- 1 root wheel 6236 12 янв 09:48 .depend lrwxr-xr-x 1 root wheel 38 12 янв 09:48 @ -> /home/devel/freebsd/base/stable/10/sys -rw-r--r-- 1 root wheel 19720 12 янв 09:50 if_gif.o lrwxr-xr-x 1 root wheel 52 12 янв 09:48 machine -> /home/devel/freebsd/base/stable/10/sys/amd64/include lrwxr-xr-x 1 root wheel 69 12 янв 09:48 opt_inet.h -> /usr/obj/home/devel/freebsd/base/stable/10/sys/LINT-NOINET/opt_inet.h lrwxr-xr-x 1 root wheel 70 12 янв 09:48 opt_inet6.h -> /usr/obj/home/devel/freebsd/base/stable/10/sys/LINT-NOINET/opt_inet6.h lrwxr-xr-x 1 root wheel 73 12 янв 09:48 opt_mrouting.h -> /usr/obj/home/devel/freebsd/base/stable/10/sys/LINT-NOINET/opt_mrouting.h lrwxr-xr-x 1 root wheel 50 12 янв 09:48 x86 -> /home/devel/freebsd/base/stable/10/sys/x86/include # cat opt_inet.h #define TCP_OFFLOAD 1 #define TCP_SIGNATURE 1
Comment Actions
I'm not surprised it works, but this is orthogonal to what is trying to be accomplished here. I'm merely trying to unbreak if_gif(4) with MK_INET == no. No more, no less.
opt_inet.h is defined in 52 Makefiles, but only 5 add TCP_OFFLOAD, and none of them add TCP_SIGNATURE:
$ grep -A 3 -lr opt_inet.h: sys/modules/ sys/modules/patm/Makefile sys/modules/sfxge/Makefile sys/modules/igb/Makefile sys/modules/fatm/Makefile sys/modules/ip_mroute_mod/Makefile sys/modules/arcnet/Makefile sys/modules/sppp/Makefile sys/modules/ixlv/Makefile sys/modules/pflog/Makefile sys/modules/if_gif/Makefile sys/modules/ipdivert/Makefile sys/modules/mlx4ib/Makefile sys/modules/ipoib/Makefile sys/modules/firewire/fwip/Makefile sys/modules/virtio/network/Makefile sys/modules/carp/Makefile sys/modules/if_tap/Makefile sys/modules/pf/Makefile sys/modules/netgraph/ipfw/Makefile sys/modules/netgraph/gif/Makefile sys/modules/netgraph/iface/Makefile sys/modules/if_bridge/Makefile sys/modules/hatm/Makefile sys/modules/an/Makefile sys/modules/wlan/Makefile sys/modules/if_faith/Makefile sys/modules/vmware/vmxnet3/Makefile sys/modules/smbfs/Makefile sys/modules/if_stf/Makefile sys/modules/mlxen/Makefile sys/modules/cxgb/tom/Makefile sys/modules/cxgb/iw_cxgb/Makefile sys/modules/cxgb/cxgb/Makefile sys/modules/if_lagg/Makefile sys/modules/ixl/Makefile sys/modules/ipfw/Makefile sys/modules/if_tun/Makefile sys/modules/em/Makefile sys/modules/en/Makefile sys/modules/snc/Makefile sys/modules/lmc/Makefile sys/modules/nfsclient/Makefile sys/modules/if_gre/Makefile sys/modules/mthca/Makefile sys/modules/if_disc/Makefile sys/modules/pfsync/Makefile sys/modules/nfscl/Makefile sys/modules/if_ef/Makefile sys/modules/cxgbe/if_cxgbe/Makefile sys/modules/cxgbe/iw_cxgbe/Makefile sys/modules/cxgbe/tom/Makefile sys/modules/mlx4/Makefile $ for i in `grep -A 3 -lr opt_inet.h: sys/modules/`; do grep -l TCP_OFFLOAD $i; done sys/modules/cxgb/tom/Makefile sys/modules/cxgb/iw_cxgb/Makefile sys/modules/cxgb/cxgb/Makefile sys/modules/cxgbe/if_cxgbe/Makefile sys/modules/cxgbe/iw_cxgbe/Makefile sys/modules/cxgbe/tom/Makefile $ for i in `grep -A 3 -lr opt_inet.h: sys/modules/`; do grep -l TCP_SIGNATURE $i; done
Comment Actions
I'm for improvement, but I'd (again) rather stick with how it has historically evolved in head and use cat in this commit to be consistent with the other code and how things evolved historically instead of inventing another mechanism, which will confuse someone potentially that follows history of the code and might apply patches in series assuming that it "just works".
Comment Actions
So what's wrong with just MFCing the changes I made to make current work without all this crazy?
Comment Actions
I feel you don't understand what I'm trying to say. These options (TCP_SIGNATURE and TCP_OFFLOAD) by themselves don't needed to build if_gif(4) module. They are orthogonal to it.
When you build the LINT-NOINET kernel where all these options are included, opt_inet.h will be generated and it will not empty and 'cat opt_inet.h' will give not empty result. Thus even when the kernel has no INET support, module that builded together with it should be build with INET.
The question is why it works and does it work as expected? Probably I just understood how it works wrongly :)
Anyway I have no objection against this. If it works for you, feel free to commit :)
Comment Actions
I didn't have the time to do this right at the moment, but I will setup another branch and pull in your changes and bz's changes to get things working.
Part of the pain stems from src.opts.mk and other build related changes that have a large degree of potential impact, which I haven't assessed yet.