From: Jarod Wilson <jarod@redhat.com>
Date: Wed, 23 Dec 2009 16:41:51 -0500
Subject: scsi: fix duplicate libiscsi symbol and kabi warnings
Message-id: <20091223164151.GA9151@redhat.com>
Patchwork-id: 22273
O-Subject: [RHEL5.5 PATCH] scsi: fix duplicate libiscsi symbol and kabi warnings
Bugzilla:

On Tue, Dec 22, 2009 at 08:18:56PM -0600, Mike Christie wrote:
> Jarod Wilson wrote:
> >On Mon, Dec 21, 2009 at 03:34:02PM -0500, Jarod Wilson wrote:
> >>[RHEL5.5 PATCH] scsi: fix duplicate libiscsi symbol warnings
> >>
> >>When libiscsi2 support was added to the RHEL5 kernel, there were
> >>some changes made to the existing libiscsi code, to add foo*iscsi1
> >>bits here and there. For the most part, nothing actually requests
> >>any of them that way, so far as I can see, and all the stuff using
> >>libiscsi2 requests *iscsi2 explicitly. At present, the changes
> >>made to the libisci1 bits result in build warnings about duplicate
> >>symbols in libiscsi.ko and libiscsi1.ko. I believe the simple fix
> >>is to only build libiscsi.ko, but I could be missing something...
> >>Also, why this didn't start showing up until very recently is
> >>currently beyond me, since the patch that created the dupes looks
> >>like it was in 5.4...
> >>
> >>Anyhow, this patch is a more complete version of what I posted
> >>in reply to the kvm/iscsi kabi patch submitted by jcm, which
> >>further reduces unnecessary duplication.
> >
> >Bah. Self-NAK. Local git tree build was okay, rpm build blew up due to
> >missing symbols that are on the whitelist. Looks like the original version
> >that only drops the duplicate libiscsi1.ko actually works though, at least
> >from an "it builds, and the dupe warnings are gone" standpoint.
> >
>
> ACK if this is ok. There is nothing in RHEL or in upstream that uses
> liiscsi1. It is only KABId due to me not understanding the process
> at the time and not requesting for it to be dropped from the KABI.
>
> There was actually one driver using the scsi_transport_iscsi1
> symbols. It was a brodacom distributed bnx2i driver. We merged the
> bnx2i driver in 5.4, so people just use that now. And, Broadcom now
> only supports the bnx2i driver we ship so there is no point in
> keeping the scsi_transport_iscsi1.

Dropping scsi_transport_iscsi1 looks a bit more involved than I'm prepared
to handle at the moment, since it would require quite a bit of code
shuffle (but maybe its just a matter of reverting key chunks of the patch
that added *iscsi1 bits in the first place). The following patch resolves
both the duplicate symbol build warnings and the kabi symbol move warnings.
It merges my v2 patch with the necessary bits of Jon's kvm kabi rfc patch
that included these symbol moves and a bit more.


diff --git a/drivers/scsi/Makefile b/drivers/scsi/Makefile
index de72583..05e4bd8 100644
--- a/drivers/scsi/Makefile
+++ b/drivers/scsi/Makefile
@@ -40,8 +40,7 @@ obj-y				+= device_handler/
 obj-$(CONFIG_LIBFC)		+= libfc/
 obj-$(CONFIG_FCOE)		+= fcoe/
 obj-$(CONFIG_FCOE_FNIC)		+= fnic/
-obj-$(CONFIG_LIBISCSI1)		+= libiscsi1.o
-libiscsi1-objs			:= libiscsi.o
+obj-$(CONFIG_LIBISCSI1)		+= libiscsi.o
 obj-$(CONFIG_ISCSI_TCP) 	+= libiscsi2.o	libiscsi_tcp.o iscsi_tcp.o
 obj-$(CONFIG_INFINIBAND_ISER) 	+= libiscsi2.o
 obj-$(CONFIG_SCSI_AMIGA7XX)	+= amiga7xx.o	53c7xx.o