Bug 429

Summary: error in compiling gmp-nativesdk for qemumips architecture on x86_64 host
Product: [Build System, Metadata & Runtime] Meta-yocto Reporter: Lianhao Lu <lianhao.lu>
Component: meta-yoctoAssignee: Saul Wold <sgw>
Status: VERIFIED OBSOLETE QA Contact:
Severity: normal    
Priority: Low CC: jessica.zhang, ke.yu, kevin.tian, liping.ke, poky.bs.watcher, poky.watcher, richard.purdie, sgw
Version: 0.9   
Target Milestone: 1.0 M4   
Hardware: x86   
OS: mips   
Whiteboard:
OS type for building Yocto: --- Type of Regression: ---
Verified: Documentation change: ---
Attachments:
Description Flags
failure log none

Description Lianhao Lu 2010-10-11 18:58:02 UTC
Created attachment 57 [details]
failure log

Using the poky-contrib/builds/milestone branch, when building the meta-toolchain-sdk target for qemumips architecture on a x86_64 host(Ubuntu 10.04), the gmp-natviesdk package failed.
Comment 1 Kevin Tian 2010-10-11 21:52:12 UTC
I tried building gmp-nativesdk directly for qemumips, there's no error. My build host is 32bit, while Lianhao's is 64bit which may be the clue.
Comment 2 Jessica 2010-10-11 22:19:37 UTC
Is there another 64bit machine that we can use to try to confirm?
Comment 3 Qing He 2010-10-12 01:18:35 UTC
This only happens when some older version of SDK (i.e. /opt/poky) is installed. If there is no /opt/poky directory, the build will pass.

So it's kind of host contamination specifically for nativesdk.
Comment 4 Qing He 2010-10-12 01:38:50 UTC
(In reply to comment #3)
> This only happens when some older version of SDK (i.e. /opt/poky) is installed.
> If there is no /opt/poky directory, the build will pass.
> 
> So it's kind of host contamination specifically for nativesdk.

details:

gmp builds and runs several binaries in the build process, which are all built with $(CC_FOR_BUILD), that's take gen-fac_ui for example

However, if /opt/poky/ exists and has an older version of sdk, these binaries may link to it instead of building sysroot, with the following test, it's clear.

---
1. $ ldd gen-fac_ui
     linux-vdso.so.1 => (...)
     libc.so.6 => ......
     /opt/poky/sysroots/x86_64-pokysdk-linux/lib/ld-linux-x86-64.so.2 => ....

2. $ ./gen-fac_ui    # parameter omitted
     Segmentation fault

3. $ /opt/poky/sysroots/x86_64-pokysdk-linux/lib/ld-linux-x86-64.so.2 ./gen-fac_ui
     Segmentation fault

4. $ /lib/ld-linux-x86-64.so.2 ./gen-fac_ui
     (passed)

5. $ /tmp/sysroots/x86_64-nativesdk-pokysdk-linux/opt/poky/sysroots/x86_64-pokysdk-linux/lib/ld-linux-x86-64.so.2 ./gen-fac_ui
     (passed)
Comment 5 Jessica 2010-10-13 11:53:44 UTC
So any update on this bug? My build went through with the exact setup except mine's a 32bit machine, and my /opt/poky maybe different from Lianhao's.  But base on Qing's comments, it should be generic issue if there's an older version of /opt/poky?  

Any, since it's a high priority bug, can we have some update or conclusion regarding it?
Comment 6 Qing He 2010-10-13 18:17:57 UTC
(In reply to comment #5)
> So any update on this bug? My build went through with the exact setup except
> mine's a 32bit machine, and my /opt/poky maybe different from Lianhao's.  But
> base on Qing's comments, it should be generic issue if there's an older version
> of /opt/poky?  

Yes, this is a generic issue (host contamination of ), but the effective damage is much more limited than previously thought.

An older /opt/poky doesn't necessarily crash the build, it doesn't crash on my host for example, the crash observed may have something to do with glibc->eglibc transition.

> Any, since it's a high priority bug, can we have some update or conclusion
> regarding it?

Since its impact is limited, I think it's OK to lower the priority.
Comment 7 Ke Yu 2010-12-08 22:39:43 UTC
set to low since its impact is limited
Comment 8 Saul Wold 2011-06-30 23:28:33 UTC
Just tested this with 1.1 Head and did not see an issue, I am going to close this bug, if you see this issue again, please re-open.
Comment 9 Lianhao Lu 2011-08-24 20:33:58 UTC
Verified in 1.1 head build, not seen this error again.