[mythtv] Problem with commit 9506b88 2017/10/09 : Cleanup gcc and clang warnings generated by -Wextra switch.

Stuart Auchterlonie stuarta at squashedfrog.net
Wed Oct 25 08:30:14 UTC 2017


On 25/10/17 09:28, Paul Harrison wrote:
> On 25/10/17 05:29, David Hampton wrote:
> 
>> Hi Peter,
>>
>> I was afraid you were going to say that it was that cleanup commit.
>> That was an aggregation of 124 smaller commits over probably six
>> months.  Given that Myth works on your AMD system and my x86 systems,
>> my guess would be that it is probably in a change to one of the openmax
>> files.
>>
>> I've looked at the changes to files with 'omx' in the name and nothing
>> stands out to me as possibly being a problem.  There appear to be only
>> four patches that modify five 'omx' files.  I've uploaded those patches
>> to http://love2code.net/mythtv/ if you want to try reverting them and
>> see if that fixes the crashing.
>>
>> I have a RPi Model B+ v1.2 that was given to me but I've never powered
>> it on.  Would this be sufficient for me to use to try and bisect the
>> problem?  I'd probably need some help in getting the RPi set up.  (Is
>> it possible to cross-compile for AMD on an x86 system, or would I need
>> to run native compiles on the RPi itself?)
>>
>> I didn't realize that git didn't push all my intermediate nodes when I
>> pushed a merge, but that makes sense.  I will certainly write merge
>> notes more carefully going forward.
>>
>> David
>>
>> P.S.  I could also revert the aggregate patch, and then rebase and
>> reapply the 124 individual patches.  That would make the individual
>> changes visible to everyone.
>>
>>
>>
>> On Tue, 2017-10-24 at 22:38 -0400, Peter Bennett wrote:
>>> Hi David
>>>
>>> The above mentioned commit is a merge of hundreds of changes. It is
>>> causing memory corruption errors on Raspberry Pi, consistently every
>>> time you try to play a recording. You get one of these errors and a
>>> core
>>> dump:
>>>
>>> corrupted size vs prev size
>>> smallbin double link list corrupted
>>> free(): invalid pointer
>>>
>>> These occur in malloc and free at different places each time. AMD64
>>> systems are not affected although the codebase is the same.
>>>
>>> Compiling from aea96ba which is the immediately prior commit, is
>>> fine.
>>> Compiling from 9506b88 which is immediately after that commit, gives
>>> those errors, as does compiling from the latest.
>>>
>>> I cannot compile any of the intermediate commits between those two,
>>> because they are off your repository, do not contain the Raspberry
>>> Pi
>>> Stretch changes, and thus fail compilation.
>>>
>>> Do you know which change may be causing this problem? We may have to
>>> roll back this set of changes.
>>>
>>> Peter
>>>
>>>
> 
> I could do another coverity build if you think it may help. The last run
> did pickup a use after free but I think that was in the crystalhd
> decoder which no one cares about.
> 
> Paul H.

Lets try that, I'll also try doing a valgrind on my dev laptop.
Might not show anything due to not using the same code path's but
it's worth a try


Regards
Stuart



More information about the mythtv-dev mailing list