From mboxrd@z Thu Jan  1 00:00:00 1970
From: Wolfgang Grandegger <wg@grandegger.com>
Date: Fri, 30 Jun 2017 11:14:23 +0200
Subject: [Buildroot] [RFC PATCH v3 00/10] Make the SDK relocatable
In-Reply-To: <1cc2e254-a03b-f8e1-9e6f-bd250c1e2496@grandegger.com>
References: <1490255693-9134-1-git-send-email-wg@grandegger.com>
 <6799ad7b-a16e-d89e-eb73-d7d91880d341@mind.be>
 <d993626e-11b0-ab21-f022-d64f6a7340b6@grandegger.com>
 <8b074253-3002-1db9-d902-a73bb47e734b@grandegger.com>
 <dec2b16c-50f5-15d2-2082-c31ca82170ec@mind.be>
 <6a236664-6cd5-c2e6-e4bc-ee5f7b6b21a1@grandegger.com>
 <311f5c7c-ad46-e1f1-4570-dfffb5605292@mind.be>
 <4243e8e2-baaf-37f5-4149-d085388f3b13@grandegger.com>
 <319491ac-9a65-f82f-892c-1e54b8df70da@mind.be>
 <b3e97095-7453-5516-0f23-e68e56932930@grandegger.com>
 <790e33dd-2285-dd7e-7203-692f2cd3686f@mind.be>
 <c9adbe27-c7b1-b28d-5c38-7c79a5da9c18@grandegger.com>
 <84234ca1-b75f-84da-80cc-83a7234d0927@mind.be>
 <d0246c22-0486-a095-0bf4-77cc14784997@grandegger.com>
 <681841f3-9d91-a2fe-d2ea-1282de1844e7@grandegger.com>
 <1cc2e254-a03b-f8e1-9e6f-bd250c1e2496@grandegger.com>
Message-ID: <b883eedb-61e7-a49a-7ec1-39e17f05327b@grandegger.com>
List-Id: <buildroot.busybox.net>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: buildroot@busybox.net

Hello, just another observation...


$ find . -name '.br_host_filelist'|xargs grep sysroot
./build/host-gcc-final-6.3.0/.br_host_filelist:usr/arm-buildroot-linux-gnueabihf/sysroot/lib/libgcc_s.so.1
./build/host-gcc-final-6.3.0/.br_host_filelist:usr/arm-buildroot-linux-gnueabihf/sysroot/usr/lib/libstdc++.a
./build/host-gcc-final-6.3.0/.br_host_filelist:usr/arm-buildroot-linux-gnueabihf/sysroot/lib/libatomic.so.1.2.0
./build/host-gcc-final-6.3.0/.br_host_filelist:usr/arm-buildroot-linux-gnueabihf/sysroot/lib/libatomic.a
./build/host-gcc-final-6.3.0/.br_host_filelist:usr/arm-buildroot-linux-gnueabihf/sysroot/usr/lib/libstdc++.so.6.0.22
./build/host-gcc-final-6.3.0/.br_host_filelist:usr/arm-buildroot-linux-gnueabihf/sysroot/lib/libatomic.la
./build/host-gcc-final-6.3.0/.br_host_filelist:usr/arm-buildroot-linux-gnueabihf/sysroot/usr/lib/libstdc++.so.6.0.22-gdb.py
./build/host-gcc-final-6.3.0/.br_host_filelist:usr/arm-buildroot-linux-gnueabihf/sysroot/lib/libgcc_s.so

GCC seems to install into "sysroot" in the "install-host" step. Not
sure if this is a problem (needs special treatment).

Wolfgang.

Am 30.06.2017 um 10:50 schrieb Wolfgang Grandegger:
> Hello,
> 
> Am 29.06.2017 um 10:02 schrieb Wolfgang Grandegger:
>> Hello Arnout,
>>
>> about rpath sanitazion of patchelf and m4...
>>
>> Am 29.06.2017 um 08:13 schrieb Wolfgang Grandegger:
>>>
>>>
>>> Am 28.06.2017 um 21:33 schrieb Arnout Vandecappelle:
>>>>
>>>>
>>>> On 28-06-17 14:36, Wolfgang Grandegger wrote:
>>>>>
>>>>>
>>>>> Am 28.06.2017 um 13:16 schrieb Arnout Vandecappelle:
>>>>>>
>>>>>>
>>>>>> On 28-06-17 12:57, Wolfgang Grandegger wrote:
>>>>>> [snip]
>>>>>>> The list is built with "comm -13 ...". While implementing that
>>>>>>> way, I realized
>>>>>>> that I need patchelf for the host as very first package. Any idea
>>>>>>> how that could
>>>>>>> achieved?
>>>>>>
>>>>>>     host-patchelf should be added to DEPENDENCIES_HOST_PREREQ so it
>>>>>> gets built
>>>>>> before all the rest. In the instrumentation hook, you should also
>>>>>> skip the
>>>>>> sanitization when the package is not in DEPENDENCIES_HOST_PREREQ.
>>>>>
>>>>> I thought all host binaries must have an rpath to <path>/usr/lib,
>>>>> which should
>>>>> be replaced with "$ORIGIN/../../../usr/lib"
>>>>
>>>>    Only if they use a library in $HOST_DIR/usr/lib - otherwise
>>>> sanitize-rpath will
>>>> remove the rpatch completely, no?
>>>
>>> Makes sense.
>>>
>>>>    DEPENDENCIES_HOST_PREREQ contains only a few packages, none of
>>>> which are linked
>>>> with any library in $HOST_DIR/usr/lib. At least I think so.
>>>
>>> In may case it only contains "patchelf", which needs rpath sanitation.
>>>
>>>>>>     That means that those packages will not get sanitized - which
>>>>>> shouldn't be
>>>>>> necessary since they anyway don't have an rpath. But perhaps it's a
>>>>>> good idea
>>>>>> then to add a readelf-based check for them that verifies they don't
>>>>>> have
>>>>>> DT_RUNPATH or DT_RPATH. The latter is just a nice-to-have, though,
>>>>>> so don't
>>>>>> worry about it at first.
>>>>>
>>>>> See above.
>>>>>
>>>>> Even with setting DEPENDENCIES_HOST_PREREQ to patchelf at the first
>>>>> place, some
>>>>> packages are required in advance to build patchelf :(.
>>>>
>>>>    Really? Which ones? patchelf has no dependencies.
>>>
>>>     host-m4-1.4.18
>>>     host-libtool-2.4.6
>>>     host-autoconf-2.69
>>>     host-automake-1.15
>>>     host-patchelf-29c085fd9d3fc972f75b3961905d6b4ecce7eb2b
>>>
>>> Here is the order in which the packages are built. Patchelf needs
>>> automake and friends. The "host-m4" binary m4 has an rpath;
>>>
>>> <buildroot-build-dir>/host/usr/lib
>>>
>>> The others have perl scripts only. Hence, we need to sanitize the
>>> "host-m4" package after patchelf is available. Not too bad.
>>
>> Neither patchelf nor m4 uses libraries from /usr/lib and therefore the
>> rpath is empty. A mistake in my scripts confused me... sorry for the noise.
> 
> I must again correct myself :(. Both binaries have
> "<buildroot-build-dir>/host/usr/lib" in their rpath after the build step and
> the sanitation will remove it:
> 
>    patching ELF file '/home/wolf/test/bdo/imx6ulpico/host/usr/bin/patchelf'
>    removing directory '/home/wolf/test/bdo/imx6ulpico/host/usr/lib' from RPATH because it does not contain needed libs
>    new rpath is ''
> 
> I have just sent out a new patch series doing the sanitation per
> package and build step. It will use "--debug" to show the sanitation
> results. I have attached the "rpath-sanitation-imx6ulpico.log" output
> to show what sanitation is doing for "imx6ulpico_defconfig" with
> "glibc", "sudo" and "pulseaudio". Some quick observations:
> 
> - It removes "/usr/lib" from most host binaries because it does not
>    contain needed libs. It uses "$ORIGIN/../../usr/lib" otherwise.
> 
> - It clears the RPATH for most target binaries. Only a few remain,
>    e.g. for the "sudo" and "pulseaudio" package (not pointing to
>    "/lib" or "/usr/lib".
> 
> Concerning that topic, I also realized that "readelf" is used with
> "regexp" heavily. I think using "patchelf" for that purpose would be
> faster.
> 
> Then happy hacking at the Summer Camp.
> 
> Wolfgang.
> 
> 
> 
> _______________________________________________
> buildroot mailing list
> buildroot at busybox.net
> http://lists.busybox.net/mailman/listinfo/buildroot
>