From mboxrd@z Thu Jan  1 00:00:00 1970
From: David Miller <davem@davemloft.net>
Subject: Re: umh build...
Date: Tue, 05 Jun 2018 19:42:48 -0400 (EDT)
Message-ID: <20180605.194248.777320854908313205.davem@davemloft.net>
References: <20180605.170359.357501154106614131.davem@davemloft.net>
        <20180605222932.k4nmh23ggf4pnsft@ast-mbp>
Mime-Version: 1.0
Content-Type: Text/Plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Cc: netdev@vger.kernel.org
To: alexei.starovoitov@gmail.com
Return-path: <netdev-owner@vger.kernel.org>
Received: from shards.monkeyblade.net ([184.105.139.130]:43620 "EHLO
        shards.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
        with ESMTP id S1752444AbeFEXvx (ORCPT
        <rfc822;netdev@vger.kernel.org>); Tue, 5 Jun 2018 19:51:53 -0400
In-Reply-To: <20180605222932.k4nmh23ggf4pnsft@ast-mbp>
Sender: netdev-owner@vger.kernel.org
List-ID: <netdev.vger.kernel.org>

From: Alexei Starovoitov <alexei.starovoitov@gmail.com>
Date: Tue, 5 Jun 2018 15:29:34 -0700

> On Tue, Jun 05, 2018 at 05:03:59PM -0400, David Miller wrote:
>> 
>> Alexei, I tried to build bpfilter on sparc64 and it shows that
>> CONFIG_OUTPUT_FORMAT is an x86 specific Kconfig value.
>> 
>> And, even if I added it, it's not clear what it should even be set to.
>> 
>> Right now, for example, my userland builds default to 32-bit sparc
>> even though I'm building a 64-bit kernel.
>> 
>> I think what ends up happening has to be in some way in response to
>> what kind of "native" binaries HOSTCC is actually building.
> 
> I guess you mean native-ness not of HOSTCC, but what target rootfs will be.

Target rootfs is where I am building the kernel in this case.

Kernel running is 64-bit, most of userland is 32-bit.  Kernel is built
explicitly with "-m64" but gcc by default is building 32-bit.

> I guess it's probably equal to 32 or 64-bitness of the kernel from
> CC.

Nope, they are and can be different, even natively.

> I completely forgot that I need to fix CONFIG_OUTPUT_FORMAT.
> On my arm64 I just did
> export CONFIG_OUTPUT_FORMAT=elf64-littleaarch64
> before doing 'make'.
> I can do a hack to extract it from objdump
> similar to the hack I do for '-B ...'
> Like the patch below...
> and it works on x64 too, but I'm not sure about sparc.
 ...
> @@ -21,7 +21,7 @@ endif
>  # which bpfilter_kern.c passes further into umh blob loader at run-time
>  quiet_cmd_copy_umh = GEN $@
>        cmd_copy_umh = echo ':' > $(obj)/.bpfilter_umh.o.cmd; \
> -      $(OBJCOPY) -I binary -O $(CONFIG_OUTPUT_FORMAT) \
> +      $(OBJCOPY) -I binary -O `$(OBJDUMP) -f $<|grep format|cut -d' ' -f8` \
>        -B `$(OBJDUMP) -f $<|grep architecture|cut -d, -f1|cut -d' ' -f2` \
>        --rename-section .data=.init.rodata $< $@

Yeah this doesn't do it in the 64-bit kernel 32-bit userspace scenerio.

We build UMH which ends up being a 32-bit ELF binary, then we later
try to link it:

  ld -m elf64_sparc   -r -o net/bpfilter/bpfilter.o net/bpfilter/bpfilter_kern.o net/bpfilter/bpfilter_umh.o ; scripts/mod/modpost net/bpfilter/bpfilter.o
ld: sparc:v8plusb architecture of input file `net/bpfilter/bpfilter_umh.o' is incompatible with sparc:v9 output

But your patch is a step in the right direction because it gets us
away from depending upon CONFIG_OUTPUT_FORMAT.