[oe] [meta-browser] Problem building chromium

[Peter A. Bigot]

On 11/02/2014 07:41 AM, Gary Thomas wrote:
> On 2014-10-31 21:36, Peter A. Bigot wrote:
>> On 10/31/2014 05:54 PM, Gary Thomas wrote:
>>> I just tried to build chromium on the Raspberry-Pi.  All repos
>>> are up to date as of 2014-10-31.  Here's the error I got:
>>>
>>> | lib/libcontent.so: undefined reference to `WebRtc_GetCPUFeaturesARM'
>>> | collect2: error: ld returned 1 exit status
>>> | ninja: build stopped: subcommand failed.
>>> | WARNING: 
>>> /home/local/rpi-latest_2014-10-30/tmp/work/armv6-vfp-amltd-linux-gnueabi/chromium/38.0.2125.101-r0/temp/run.do_compile.10091:1 
>>> exit 1 from
>>> |   ninja -C 
>>> /home/local/rpi-latest_2014-10-30/tmp/work/armv6-vfp-amltd-linux-gnueabi/chromium/38.0.2125.101-r0/chromium-38.0.2125.101/out/Release 
>>> chrome chrome_sandbox
>>> | ERROR: Function failed: do_compile (log file is located at
>>> /home/local/rpi-latest_2014-10-30/tmp/work/armv6-vfp-amltd-linux-gnueabi/chromium/38.0.2125.101-r0/temp/log.do_compile.10091) 
>>>
>>> ERROR: Task 592 
>>> (/home/local/poky-latest/meta-browser/recipes-browser/chromium/chromium_38.0.2125.101.bb, 
>>> do_compile) failed with exit code '1'
>>>
>>> Note: I was able to build this same recipe just fine for my
>>> i.MX6 platforms, so it must be something to do with armv6 vs
>>> armv7a/cortex.  I've not found it yet but thought someone might
>>> have ideas.
>>>
>>> Thanks
>>>
>>
>> As noted at 
>> http://lists.openembedded.org/pipermail/openembedded-core/2014-August/096568.html 
>> and related messages, raspi configures (or at that time configured) 
>> gcc to build for
>> armv6 but it's actually armv6zk.  It's entirely possible chromium 
>> doesn't support armv6 without armv6k features being enabled. My best 
>> guess is meta-raspberrypi needs to be fixed,
>> which might require some enhancements in oe-core's gcc.
>
> While what you say may be true or even relevant, chromium
> used to build and run fine for the RaspberryPi.  That said,
> I don't see how an instruction set choice/mixup explains
> a missing symbol.

My thought was something titled WebRtc_GetCPUFeaturesARM might be 
related to the ARM features selected by the compiler.

But a little googling suggests there are plenty of alternative causes 
that don't warrant my instinctual "it's not gcc's fault!" response.

https://github.com/RobertCNelson/pkg-chromium/blob/master/patches/arm-webrtc-fix.patch 
looks relevant.

Possibly related to that is the following at the bottom of 
meta-browser's README:

TODO
----
[...]
- fix Chromium support other platforms than armv7

thought that TODO has been there since 2012, so perhaps that's not 
relevant either.

At any rate, it's not gcc's fault.  You might try reverting meta-browser 
back to get a chromium version that used to work.

Peter

>
> Following up, I stubbed out this missing symbol.  Now I can
> get it to link, but it gets an illegal instruction immediately :-(
> I tried debugging using GDB, but for some reason the library that
> it fails in has no symbols:
>
> (gdb) tar rem 192.168.1.150:9000
> Remote debugging using 192.168.1.150:9000
> Reading symbols from 
> tmp/work/raspberrypi-amltd-linux-gnueabi/amltd-x11-image/1.0-r0/rootfs/lib/ld-linux.so.3...Reading 
> symbols from 
> tmp/work/raspberrypi-amltd-linux-gnueabi/amltd-x11-image/1.0-r0/rootfs/lib/.debug/ld-2.20.so...done.
> done.
> Loaded symbols for 
> tmp/work/raspberrypi-amltd-linux-gnueabi/amltd-x11-image/1.0-r0/rootfs/lib/ld-linux.so.3
> 0xb6fd8d10 in _start () from 
> tmp/work/raspberrypi-amltd-linux-gnueabi/amltd-x11-image/1.0-r0/rootfs/lib/ld-linux.so.3
> (gdb) c
> Continuing.
>
> Program received signal SIGILL, Illegal instruction.
> 0xb61f60ba in ?? ()
>    from 
> tmp/work/raspberrypi-amltd-linux-gnueabi/amltd-x11-image/1.0-r0/rootfs/usr/lib/chrome/libbase.so
> (gdb) x/10i 0xb61f60a8
>    0xb61f60a8:  bkpt    0x006c
>    0xb61f60aa:  movs    r6, r0
>    0xb61f60ac:  ldr     r3, [pc, #12]   ; (0xb61f60bc)
>    0xb61f60ae:  ldr     r2, [pc, #16]   ; (0xb61f60c0)
>    0xb61f60b0:  add     r3, pc
>    0xb61f60b2:  movs    r1, #0
>    0xb61f60b4:  ldr     r0, [r3, r2]
>    0xb61f60b6:  b.w     0xb61f24d4
> => 0xb61f60ba:  nop
>    0xb61f60bc:  b.n     0xb61f61d8
> (gdb) x/x 0xb61f60ba
> 0xb61f60ba:     0xe08cbf00
>
> Anyone have any ideas what might be illegal here (and how to
> figure out where this code is in the source tree)?
>
> This platform seems a bit shaky when using Poky/Yocto :-(
> I tried other web browsers without much joy - firefox seems
> to work OK but midori also dies quickly in its javascript
> engine...
>