[lfs-dev] kernel messages during gcc tests

Pierre Labastie pierre.labastie at neuf.fr
Wed Feb 3 09:33:50 PST 2016


On 03/02/2016 15:08, William Harrington wrote:
> On Wed, February 3, 2016 04:36, Bruce Dubbs wrote:
>> I just updated several packages in LFS and did a complete LFS build
>> including all Chapter 6 tests.  One thing I noticed that I haven't seen
>> before is a ton of error messages in /var/log/kern.log.  The look like:
>>
>> Feb  2 20:52:05 lfs78 kernel: [1888937.495544] lib-42.exe[6559]: segfault
>> at 8 ip 00007f63a68c1fbe sp 00007fffd02bbde0 error 4 in
>> libgomp.so.1.0.0[7f63a68ab000+1f000]
>>
>> In addition to segfaults, there are some messages about 'trap invalid
>> opcode' in overflow-add-4.exe.
>>
>>  From the time stamps, I traced these to the gcc testuite.  The test log
>> does not show anything unusual.  There are two unexpected failures in
>> libstdc++:
>>
>> FAIL: experimental/filesystem/iterators/directory_iterator.cc
>> FAIL: experimental/filesystem/iterators/recursive_directory_iterator.cc
>>
>> But I've seen those before.
>>
>> Overall the build seems to have been normal.
>>
>> A couple of things to consider is that it may be a HW issue on my system.
>>   It's a new Skylake i5, so gcc may be making the wrong assumptions about
>> instruction set.  I can check this out on an Haswell i7 to see if it
>> happens there.
>>
>> Overall, these may just artifacts in the gcc test procedure, but has
>> anyone else seen anything like this?
>>
>>
>>    -- Bruce
> Hello Bruce,
> 
> I haven't checked kernel messages during GCC testuite builds, but I have
> noticed segfaults via kernel messages when testing GLIBC in the past. Did
> you get any during the GLIBC tests, too?
> 
> I will usually run dmesg at the end of the build to see if anything
> strange occurred throughout the build.
> 
> Sincerely,
> 
> William Harrington
> 
> 

Maybe it could be included into jhalfs... Run dmesg after each build. I know I
have seen kernel messages before, but bever bothered to investigate. The only
one I know the reason for is PHP's configure (in BLFS), where some test
program explicitely wants to segfault...

There are also a lot of kernel messages during LVM tests. I have added a
suggestion to disable them during the build on the page in BLFS...

Pierre


Pierre


More information about the lfs-dev mailing list