svn commit: r249410 - in head/sys: amd64/conf arm/conf cam/ctl conf i386/conf ia64/conf modules/ctl sparc64/conf

Marius Strobl marius at alchemy.franken.de
Tue Jul 30 21:41:02 UTC 2013


On Tue, Jul 30, 2013 at 10:40:49PM +0200, Edward Tomasz Napiera?a wrote:
> Wiadomo?? napisana przez Marius Strobl <marius at alchemy.franken.de> w dniu 30 lip 2013, o godz. 22:28:
> > On Tue, Jul 30, 2013 at 09:57:21PM +0200, Edward Tomasz Napiera?a wrote:
> >> Wiadomo?? napisana przez Marius Strobl <marius at alchemy.franken.de> w dniu 29 lip 2013, o godz. 22:38:
> >>> On Fri, Apr 12, 2013 at 04:25:03PM +0000, Edward Tomasz Napierala wrote:
> >>>> Author: trasz
> >>>> Date: Fri Apr 12 16:25:03 2013
> >>>> New Revision: 249410
> >>>> URL: http://svnweb.freebsd.org/changeset/base/249410
> >>>> 
> >>>> Log:
> >>>> Remove ctl(4) from GENERIC.  Also remove 'options CTL_DISABLE'
> >>>> and kern.cam.ctl.disable tunable; those were introduced as a workaround
> >>>> to make it possible to boot GENERIC on low memory machines.
> >>>> 
> >>>> With ctl(4) being built as a module and automatically loaded by ctladm(8),
> >>>> this makes CTL work out of the box.
> >>>> 
> >>> 
> >>> Uhm, shouldn't r249328 and the above change be MFCed to stable/9 in
> >>> order to reduce the default memory footprint there?
> >> 
> >> Yup, my bad.  I'm waiting for reply from re@ whether it's a good thing
> >> to do at this point of the release process.
> >> 
> > 
> > Yeah, I saw that request and was looking into what it would actually
> > take to remove ctl(4) from GENERICs in stable/9. Unfortunately, it's
> > not exactly as straight forward as MFCing the two above revisions 1:1.
> > For one, IMO kern.cam.ctl.disable shouldn't be removed there in case
> > someone uses an old kernel configuration file having "device ctl" but
> > relies on that tunable to work. That's no big deal, though, and makes
> > the merge even less intrusive. However, the real problem is that ctl.ko
> > currently is globally disabled in stable/9 as it doesn't build with PAE/
> > XEN. Actually, the compiler is right to complain as a 32-bit void-pointer
> > is casted to a 64-bit integer in that case. What I don't get so far is
> > why on earth this doesn't break compilation in head ...
> 
> Damn.  Now that you mention it... yeah, it was me who disconnected it,
> in r249530.  I had no idea what's going on either.
> 

Well, could be a clang vs. GCC thing but I currently can't build an
i386 world from head with GCC due to yet another breakage in order to
verify that *sigh*. Anyway, it seems support for bus address still is
to be properly implemented in ctl(4) so it doesn't make much sense to
fiddle with making that part compile on all architectures. The bottom
line probably is that the least intrusive thing to do is to request an
approval for a direct commit to stable/9 reverting r249530 but disabling
ctl.ko for XEN with something like the below patch. I did run that
through a universe build and it didn't break anything.

Marius

Index: i386/conf/XEN
===================================================================
--- i386/conf/XEN	(revision 253803)
+++ i386/conf/XEN	(working copy)
@@ -7,7 +7,7 @@ cpu		I686_CPU
 ident		XEN
 
 makeoptions	DEBUG=-g		# Build kernel with gdb(1) debug symbols
-makeoptions	WITHOUT_MODULES="aha ahb amd cxgb dpt drm drm2 hptnr hptmv ida malo mps mwl nve rdma sound sym trm xfs"
+makeoptions	WITHOUT_MODULES="aha ahb amd ctl cxgb dpt drm drm2 hptnr hptmv ida malo mps mwl nve rdma sound sym trm xfs"
 
 options 	SCHED_ULE		# ULE scheduler
 options 	PREEMPTION		# Enable kernel thread preemption
Index: modules/Makefile
===================================================================
--- modules/Makefile	(revision 253803)
+++ modules/Makefile	(working copy)
@@ -76,6 +76,7 @@ SUBDIR=	\
 	${_cryptodev} \
 	${_cs} \
 	${_ctau} \
+	ctl \
 	${_cxgb} \
 	cxgbe \
 	${_cyclic} \



More information about the svn-src-all mailing list