svn commit: r281643 - head/sys/net

[Peter Wemm]

On Sunday, May 03, 2015 10:15:13 PM Peter Wemm wrote:
> On Sunday, May 03, 2015 09:52:08 PM Peter Wemm wrote:
> > On Sunday, May 03, 2015 09:31:01 PM Peter Wemm wrote:
> > > On Friday, April 17, 2015 06:39:16 AM Gleb Smirnoff wrote:
> > > > Author: glebius
> > > > Date: Fri Apr 17 06:39:15 2015
> > > > New Revision: 281643
> > > > URL: https://svnweb.freebsd.org/changeset/base/281643
> > > > 
> > > > Log:
> > > >   Bring in if_types.h from projects/ifnet, where types are
> > > >   defined in enum.
> > > > 
> > > > Modified:
> > > >   head/sys/net/if_types.h
> > > 
> > > I'm sorry, but this breaks the freebsd.org cluster.
> > > 
> > > 281642 works
> > > 281643 fails
> > > head with 281643 reverted works
> > > 
> > > IPv6 doesn't work on systems with pf, carp and multiple interfaces.  I
> > > have
> > > nothing more specific than that.
> > 
> > The kernel is smaller with the enums:
> >           text     data       bss           dec             hex
> > 
> > 11438123   730686   2998048   15166857   0xe76d89
> > 11437787   730686   2998048   15166521   0xe76c39
> > 
> > The first is with the #defines, the second is enums.  The enums caused 336
> > bytes less code to be generated, so presumably the enums enabled some
> > "optimization" that wasn't expected.
> 
> More data; looking at size *.o in the kernel compile obj directory to see
> what changes size with enum vs #define:
> @@ -470,10 +470,10 @@
>      1158     124        0     1282     0x502   imgact_shell.o
>     11617     160       36    11813    0x2e25   in.o
>       343       0        0      343     0x157   in4_cksum.o
> -   22155       0        0    22155    0x568b   in6.o
> +   22187       0        0    22187    0x56ab   in6.o
>      2084       0        0     2084     0x824   in6_cksum.o
>      1155     200        0     1355     0x54b   in6_gif.o
> -    5208      12       16     5236    0x1474   in6_ifattach.o
> +    5224      12       16     5252    0x1484   in6_ifattach.o
>     21933     864        0    22797    0x590d   in6_mcast.o
>      6860       0        0     6860    0x1acc   in6_pcb.o
>      2627    6960      248     9835    0x266b   in6_proto.o
> @@ -750,8 +750,8 @@
>     25424     452        0    25876    0x6514   mvs.o
>         0      24        0       24      0x18   mvs_if.o
>      5989     268        0     6257    0x1871   mvs_pci.o
> -   17111     388       72    17571    0x44a3   nd6.o
> -   15251      92       48    15391    0x3c1f   nd6_nbr.o
> +   17287     388       72    17747    0x4553   nd6.o
> +   15366      92       48    15506    0x3c92   nd6_nbr.o
>     16318      66       16    16400    0x4010   nd6_rtr.o
>       466     196        8      670     0x29e   nehemiah.o
>      7844    2412     2036    12292    0x3004   netisr.o
> 
> The size changes are confined to the ipv6 stack.

I've just been pointed to the #ifdef IFT_* fix for this, at about the same time 
I was looking at the C code.  Sigh, fixed already, I believe.


-- 
Peter Wemm - peter at wemm.org; peter at FreeBSD.org; peter at yahoo-inc.com; KI6FJV
UTF-8: for when a ' or ... just won\342\200\231t do\342\200\246
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 473 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.freebsd.org/pipermail/svn-src-head/attachments/20150503/ffbe430e/attachment.sig>