On 8/7/2016 9:58 PM, Stephen Rothwell wrote:

Hi all,

With Linus' tree, today's linux-next build (powerpc allyesconfig) failed
like this:

drivers/infiniband/sw/built-in.o:(.opd+0x1698): multiple definition of `copy_data'
drivers/infiniband/hw/built-in.o:(.opd+0xe5f8): first defined here
drivers/infiniband/sw/built-in.o:(.opd+0x1320): multiple definition of `rxe_av_from_attr'
drivers/infiniband/hw/built-in.o:(.opd+0xe280): first defined here

This looks very suspicious.  I think you are picking up two copies of
the rxe code.  One from Linus tree and one from probably Leon's tree.
The thing is, the rxe driver is an entirely software RoCE driver.  So,
after I picked it up, I moved it from drivers/infiniband/hw to
drivers/infiniband/sw where it more correctly belongs.  Leon's tree
doesn't have that move, and since you pull Leon's tree into your -next
process, with the allyesconfig you are trying to build two copies of it
and put them both statically into the kernel, hence all of the double
definitions.  Leon needs to drop his copy of the rxe driver and let the
one in Linus' tree be the only one in existence.

drivers/infiniband/sw/built-in.o:(.opd+0x18d8): multiple definition of `rxe_do_task'
drivers/infiniband/hw/built-in.o:(.opd+0xe838): first defined here

and lots of others.

Caused by commit

  8700e3e7c485 ("Soft RoCE driver")

(which was never in linux-next :-()

The code was, just not that specific commit.  I both moved the code to
infiniband/sw and also squashed it all down to one commit.  After I did
the move/squash, I got 0day and local testing to make sure I didn't
break it in the move, but I didn't get linux next as I knew it had
already been through that and a move/squash should not break anything
that linux next is testing/checking.

Leon, please drop your rxe code out of your linux-next area so we can
eliminate this issue (or replace it with the final commit I submitted).

I just disabled teh driver for today:

-- 
Doug Ledford [off-list ref]
    GPG Key ID: 0E572FDD