search

solettaproject / soletta

Soletta Project is a framework for making IoT devices. With Soletta Project's libraries developers can easily write software for devices that control actuators/sensors and communicate using standard technologies. It enables adding smartness even on the smallest edge devices.
http://solettaproject.org
Apache License 2.0
225 stars 108 forks source link

Build error with latest Quark SOC v1.2 tool chain #1941

Open lblim opened 8 years ago

lblim commented 8 years ago

Issue Description

Need the right tool chain to cross-compile Soletta image for Galileo platform and clearer instruction that whether Soletta must be built with uclibc. The instructions on the webpage https://github.com/solettaproject/soletta/wiki/Galileo-and-Galileo-Gen2-Instructions points to the older Quark SOC X1000 BSP (v1.0.1), please update to the latest BSP (v1.2).

For the v1.0.1 the SDK tool chain (http://download.intel.com/support/processors/quark/sb/quark_bsp_buildandswuserguide_329687_006.pdf) by default is uclibc, but the latest BSP (http://www.intel.com/content/dam/www/public/us/en/documents/release-notes/quark-x1000-bsp-build-sw-rel-user-guide.pdf) can output two different tool chain, for SPI image will use uclibc but DEFAULT full image will use glibc.

I tried to build with the latest toolchains (both uclibc and glibc) are hitting into the same error:

  CC   build/stage/modules/flow/gpio/gpio.o
  CC   build/stage/modules/flow/aio/aio.o
  CC   build/stage/modules/flow/converter/converter.o
  CC   build/stage/lib/common/sol-blob.o

cc1: warning: ./build/stage/include/: No such file or directory In file included from :0:0: /home/phoongst/mywork/soletta/src/lib/common/sol-missing.h:67:1: error: static declaration of 'memrchr' follows non-static declaration memrchr(const void _haystack, int c, size_t n) ^ In file included from /home/phoongst/mywork/soletta/src/lib/common/sol-missing.h:64:0, from :0: /opt/iot-devkit/1.7.2/sysroots/i586-poky-linux/usr/include/string.h:76:14: note: previous declaration of 'memrchr' was here extern void memrchr (const void s, int c, size_t _n) ^ make: ** [build/stage/lib/common/sol-blob.o] Error 1

I then tried to build the older tool chain based on the instruction, unfortunately it failed when building the tool chain

Running: git init Initialized empty Git repository in /home/phoongst/mywork/quark_sdk/1.0.1/meta-clanton_v1.0.1/poky/.git/ Running: git remote add origin git://git.yoctoproject.org/poky Running: git fetch origin --depth=3 tags/dylan-9.0.2:tags/dylan-9.0.2 remote: Counting objects: 5108, done. remote: Compressing objects: 100% (4591/4591), done. remote: Total 5108 (delta 332), reused 2736 (delta 206) Receiving objects: 100% (5108/5108), 9.78 MiB | 742.00 KiB/s, done. Resolving deltas: 100% (332/332), done. From git://git.yoctoproject.org/poky

Host OS and architecture: Ubuntu 14.04.4 LTS

Target OS and platform: Linux, Galileo Gen 2

Logs

brunobottazzini commented 8 years ago

Hello @lblim,

Clanton is an outdated build for Yocto, and we highly recommend the use of Ostro OS.

Ostro OS is a Yocto-based operating system for IoT devices.

Try reading this working-progress documentation here and follow the steps: https://github.com/solettaproject/soletta/wiki/Meta-Soletta

Feel free to contact me if you have any doubt.

mbelluzzo commented 8 years ago

Hey @lblim , Could you run 'make distclean' (or even 'git clean -dxf' if you don't have uncommitted code) before trying the instructions to see if the problem is solved?

If you check your config file you can see that CONFIG_CFLAGS is empty, which means that you probably compiled Soletta for your host before trying to compile for Galileo. After the environment setup, CFLAGS have the SDK info, so you should see that on your .config but is probably not re-generated. The suggested commands should force that.

If it works, plz let us know so we can enforce those steps in the instructions text.

lblim commented 8 years ago

Hi @brunobottazzini I am building the Soletta on its own and not with Ostro because we have not migrated to ostro OS, Ostro OS does not have some of the ingredients of our BSP. I plan to build Soletta for Apollo Lake while waiting for my team to integrate soletta into our BSP.

lblim commented 8 years ago

Hi @mbelluzzo

I tried with make distclean and git clean -dxf with a clean checkout of soletta, tried with both uclibc and glibc toolchains, both reported same error, but this time the error is different from the previous error. I tried with the v1_beta19 and master branch.

make log = http://pastebin.com/piAkGjVw .config = http://pastebin.com/GrYVqKwX config.log = http://pastebin.com/bQeznQrz strace.zip

barbieri commented 8 years ago

The logs are to worry about, it's mixing with your host system, things like $PATH, $PKG_CONFIG_PATH, $PKG_CONFIG_LIBDIR must be all changed to restrict access to your SDK.

also, you're missing lots of dependencies that are useful for a proper soletta application, like libcurl for http-client, libudev for device enumeration, libsystemd to control services and system state, openssl...

Specifically about make log (http://pastebin.com/piAkGjVw) you're trying to build micro-linux platform (instead of systemd, since you lack libsystemd) and that results in a failure since your linux/reboot.h header lack the suspend value RB_SW_SUSPEND.

uCLib is a poor choice of library as it misses lots of utilities we use in our linux port, like the printf() implementation is horrible and all. When attempting to do minimal systems, use musl-libc instead.

Last but not least, when you say "Ostro does not have your BSP", that's just a matter of adding your layer and using it with a higher priority than Ostro's. Please use the new instructions at https://ostroproject.org/documentation/howtos/building-images.html#adding-a-custom-layer-in-ostro-os

Cross compiling complex software with multiple dependencies such as soletta is harder than you expect :-)

lblim commented 8 years ago

Hi @Barbieri,

Thanks for the info, there are a few things I would like to suggest for the documentation update. It seems that it is not encourage to cross compile the soletta image due to dependencies, please update the page https://github.com/solettaproject/soletta/wiki/Galileo-and-Galileo-Gen2-Instructions accordingly with the working workflow or remove teh cross compile option.

I did try the uclibc and glibc and both hit into same prb, so should be the system issue. So looks like system is a MUST have component. The instruction actually asked to use Linux micro option and also the instruction asked to unset the $PKG_CONFIG_PATH etc.

Besides that, the build requirements (https://github.com/solettaproject/soletta/wiki/Build-requirements) may need to be updated since the libcurl, libudev are required and not optional.

My team is working on integrating the meta-soletta into APL-I yocto BSP now. I shall wait for them to enable it since that I am unable to proceed with cross-compile option.

Enabling Ostro OS for APL-I is in your roadmap now, so my team will pick up from there and add on meta layers that are missing to support our customers.

barbieri commented 8 years ago

Hi Lee Booi,

Indeed the instructions need to be improved, there are multiple tasks pending on that :-)

The libraries are optional, but recommended. You may drop them, but you must know what you're doing. Usually would drop those in order to save memory.

It's strange glibc misses the symbol in linux/reboot.h since that exists since kernel 2.6. Maybe you missed kernel headers in your libc?

lblim commented 8 years ago

Hi Gustavo,

I found the problem already. The GCC is v4.9.1 and the glibc is v2.19, this version actually has the definition of RB_SW_SUSPEND. RB_SW_SUSPEND is defined in sys/reboot.h and it is only added after v2.16 (2012 May). However, the sysroots of the SDK somehow has the older glibc header files. Therefore is a mismatch on the glibc header files version. I am not sure why this discrepancy happen yet, I will talk to quark team on that.

commit 1a09dc565db1da971ade18d9be7d9ac82646d599 Author: Andreas Jaeger aj@suse.de Date: Wed May 23 20:16:00 2012 +0200

Add more constants to Linux <sys/reboot.h>

2012-05-23  Andreas Jaeger  <aj@suse.de>
        Maximilian Attems  <max@stro.at>

    * sysdeps/unix/sysv/linux/sys/reboot.h (RB_SW_SUSPEND,
    RB_KEXEC):
    New macros.
barbieri commented 8 years ago

ok... this is the reason why Ostro exists, the previous SDK for Galileo and Edison were painful to use, weird choice of components and can't scale to add more components (the traditional YP SDK is just the compiler, while the new eSDK embeds bitbake/devtool, then allow to add more recipes)

lblim commented 8 years ago

I got the workaround for the toolchain already

Option 1 - Use the build environment SDK (created using "bitbake meta-ide-support")

Option 2 - If an SDK installer script is required (to install an SDK to a different computer) then make this simple change to meta/lib/oe/manifest.py and create the installer script again: Move PKG_TYPE_MULTILIB from last to first in the INSTALL_ORDER.

Before INSTALL_ORDER = [ PKG_TYPE_LANGUAGE, PKG_TYPE_MUST_INSTALL, PKG_TYPE_ATTEMPT_ONLY, PKG_TYPE_MULTILIB ]

After INSTALL_ORDER = [ PKG_TYPE_MULTILIB, PKG_TYPE_LANGUAGE, PKG_TYPE_MUST_INSTALL, PKG_TYPE_ATTEMPT_ONLY ]

However, I hit into another issue as below:

  CC   build/stage/modules/flow/console/console.o
  CC   build/stage/modules/flow/byte/byte.o
  CC   build/stage/modules/flow/led-7seg/led-7seg.o
  LD   build/soletta_sysroot/usr/lib/libsoletta.so

./src/lib/comms/sol-oic-server.c: In function 'res_payload_do.isra.0': ./src/lib/comms/sol-oic-server.c:239:1: internal compiler error: in propagate_rhs_into_lhs, at tree-ssa-dom.c:2913 res_payload_do(CborEncoder _encoder, ^ Please submit a full bug report, with preprocessed source if appropriate. See http://gcc.gnu.org/bugs.html for instructions. lto-wrapper: i586-poky-linux-gcc returned 1 exit status /opt/quarksdk/1.7.2/sysroots/i586-pokysdk-linux/usr/bin/i586-poky-linux/../../libexec/i586-poky-linux/gcc/i586-poky-linux/4.9.1/ld: lto-wrapper failed collect2: error: ld returned 1 exit status make: ** [build/soletta_sysroot/usr/lib/libsoletta.so] Error 1

Any idea?

barbieri commented 8 years ago

Maybe you can try to remove -ftlo usage. Not sure if we have a way to do it now, @dorileo do you know?

This GCC 4.9 is ancient, LTO was just seeing its first users... much improved in 5.x series.

dorileo commented 8 years ago

please run make with V=1 and show us the output, -flto will only be used if building with -O[0-5].

dorileo commented 8 years ago

actually the other way around, -flto will only be used if not building with -O[0-5]

barbieri commented 8 years ago

strange, usually LTO is used with optimizations... are you sure about that?

Anyway, try the DEBUG build, AFAIR it doesn't include the LTO

dorileo commented 8 years ago

@barbieri you're right, misread in the second run :(

ifneq (,$(shell echo $(COMMON_CFLAGS) | grep -e "-O[s12345\ ]"))

lblim commented 8 years ago

•Soletta version (top commit hash): v1_beta19 (commit 6a7bc3) .config : http://pastebin.com/4pGmTspJ config.log : http://pastebin.com/1rBsEw7T make log : http://pastebin.com/Bs5Yk4Yd strace.zip

lblim commented 8 years ago

Another build error with latest master branch code instead of v1_beta19, failed at make menuconfig

•Soletta version (top commit hash): master (commit 6f93cf1042) .cpnfig : NIL config.log : http://pastebin.com/EMHFYZEh make log : http://pastebin.com/ETXwvaTu make strace log (strace -ff make &> strace.log): strace.zip make strace log (strace -ff make menuconfig &> strace.log): strace2.zip strace.zip strace2.zip

mbelluzzo commented 8 years ago

This happens when you forget to unset pkg_config_path ? This is usually the case with this error.

On Tue, May 10, 2016, 03:31 LB Lim notifications@github.com wrote:

Another build error with latest master branch code instead of v1_beta19, failed at make menuconfig

•Soletta version (top commit hash): master (commit 6f93cf1 https://github.com/solettaproject/soletta/commit/6f93cf1042e4b7ec04208821f5622b566b04f407 ) .cpnfig : NIL config.log : http://pastebin.com/EMHFYZEh make log : http://pastebin.com/ETXwvaTu make strace log (strace -ff make &> strace.log): strace.zip make strace log (strace -ff make menuconfig &> strace.log): strace2.zip strace.zip https://github.com/solettaproject/soletta/files/256490/strace.zip strace2.zip https://github.com/solettaproject/soletta/files/256491/strace2.zip

— You are receiving this because you were mentioned. Reply to this email directly or view it on GitHub https://github.com/solettaproject/soletta/issues/1941#issuecomment-218070398

lblim commented 8 years ago

Nope, I have unset the PKG_CONFIG_PATH. This is the script I used to build and reproduce the prb

cd mywork/soletta git pull make distclean git clean -dfx git submodule init && git submodule update source /opt/quark_sdk/1.7.2/environment-setup-i586-poky-linux unset PYTHONHOME export TARGETCC=$CC export TARGETAR=$AR unset PKG_CONFIG_PATH make menuconfig

barbieri commented 8 years ago

likely @mbelluzzo wrote one thing to mean the other: If you set PKG_CONFIG_PATH and PKG_CONFIG_LIBDIR you can avoid the cross compilation to look at your host.

That is, if you unset (as he wrote), then you're likely using /usr/lib/pkgconfig while you'd be interested in /opt/quark_sdk/1.7.2/usr/lib/pkgconfig (or something like that), that would include dependencies for libcurl, icu, mosquitto, microhttpd and so on.

barbieri commented 8 years ago

Also, looking at your pastebins. I'd recommend to rebase to last master as we're improving some bits, I se converter/string-format.c but that's now at format/string-format.c, the warning looks harmless there.

The warning in sol-memmap-storage.c:178 is also strange, fread(&old_value, entry->size, 1, file); with old_value being uint64_t (8 bytes), then entry->size is checked to be less or equal 8 with an assert: assert(entry->size <= 8);, so the supposed error from the compiler can never be reached.

The actual build error is resulting from an internal compiler error, that is, GCC itself is crashing.

My recommendation is to skip the old GCC and try a newer toolchain with 5.3, like the one in Ostro.

mbelluzzo commented 8 years ago

Nope, I meant unset. We unset PKG_CONFIG_PATH so our host tools find the needed dependencies just fine (IOW do not look in the sdk to compile host stuff), like ncurses to display the menuconfig. And this doesn't break the cross compile because it's done through --sysroot flags everywhere and therefore should use the dir pointed at PKG_CONFIG_SYSROOT_DIR for everything related to it (iow the target dependencies), avoiding the mix of host and target dependencies. Unless something changed on yocto 1.7.2 environment script.

barbieri commented 8 years ago

then you're lucky, because it makes no sense :-D PKG_CONFIG_SYSROOT_DIR affects ${prefix} replacement.

Or pkg-config that is being used is the one provided by the SDK, usually patched to assume PKG_CONFIG_LIBDIR to the sdk installation path

mbelluzzo commented 8 years ago

make V=1 while I cross compile says I'm not having luck, just pointing things to the right stuff.

On Wed, May 11, 2016 at 2:48 PM, Gustavo Sverzut Barbieri < notifications@github.com> wrote:

then you're lucky, because it makes no sense :-D PKG_CONFIG_SYSROOT_DIR affects ${prefix} replacement.

Or pkg-config that is being used is the one provided by the SDK, usually patched to assume PKG_CONFIG_LIBDIR to the sdk installation path

— You are receiving this because you were mentioned. Reply to this email directly or view it on GitHub https://github.com/solettaproject/soletta/issues/1941#issuecomment-218536447

barbieri commented 8 years ago

nice way to try this is to install a -dev package on your host that you do not have on your target, like icu-dev or something like this. See if make reconf is detecting it on your host and not on cross-compile.

mbelluzzo commented 8 years ago

That is already the case when I compile Soletta to run on original Edison (with the original SDK), and works just fine, including an observation that "You're building the string nodes module without i18n support.", because the dependency isn't on the target, but it is on my host.

On Wed, May 11, 2016 at 3:40 PM, Gustavo Sverzut Barbieri < notifications@github.com> wrote:

nice way to try this is to install a -dev package on your host that you do not have on your target, like icu-dev or something like this. See if make reconf is detecting it on your host and not on cross-compile.

— You are receiving this because you were mentioned. Reply to this email directly or view it on GitHub https://github.com/solettaproject/soletta/issues/1941#issuecomment-218551200