mesos-reviews mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Ayanampudi Varsha <avar...@us.ibm.com>
Subject Re: Review Request 55065: Added support for s390x architecture
Date Tue, 28 Mar 2017 10:34:13 GMT


> On March 22, 2017, 2:45 p.m., Till Toenshoff wrote:
> > src/linux/ldcache.cpp
> > Line 120 (original), 120 (patched)
> > <https://reviews.apache.org/r/55065/diff/1/?file=1592794#file1592794line120>
> >
> >     Can you elaborate why this change is safe for all systems?
> 
> Ayanampudi Varsha wrote:
>     We have tried this on Big endian system where 16 bit flag was causing issue.
>     Test case LdcacheTest.Parse failed at condition ASSERT_GT(cache->size(), 1u);
as size was returned as 0. This happens because the 16 bits after the field "flags" is unused.

>     On x86_64, I have verified that this "cache->size()" is same with 16 bit and 32
bit flag.
>     Observed that the other fields in structure are all 32bit/64bit.
>     An alternative would be to make changes only specific to s390x by putting a condition.
Please suggest
> 
> Till Toenshoff wrote:
>     You may actually have found a bug in our ldcache parsing. After looking into the
glibc sources, I think this field should have been 32bit also for x86_64 instead of 16bit
- I suspect missing struct-packing for this bug to not have surfaced earlier. For confirming
this without mixing it with your platform specific docker-repository change, I would like
you to split this patch into two.
>     
>     So please add another patch that only fixes `flags` attribute size into 32bit. Please
make sure you add the following reviewers: tillt, klueska.
>     
>     Remove that 32bit change from this patch and update it according to my and Kevin's
review comments.
>     
>     Thanks a bunch for your work - this is highly appreciated.
> 
> Till Toenshoff wrote:
>     Ow, for further reference of this `flags` change, here are the internal glibc structures:
https://github.com/lattera/glibc/blob/master/sysdeps/generic/dl-cache.h#L80

As I am planning to drop the docker_registry customization change, can I resubmit removing
docker_registry change in same patch and continue in the same review board ticket. please
suggest


- Ayanampudi


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/55065/#review169708
-----------------------------------------------------------


On March 27, 2017, 9:25 a.m., Ayanampudi Varsha wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/55065/
> -----------------------------------------------------------
> 
> (Updated March 27, 2017, 9:25 a.m.)
> 
> 
> Review request for mesos, Till Toenshoff and Vinod Kone.
> 
> 
> Bugs: MESOS-6742
>     https://issues.apache.org/jira/browse/MESOS-6742
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Added support for s390x architecture
> There are 2 issues addressed in this patch:
> 1. LdcacheTest.Parse test case fails on s390x machines.
> 2. From the value of flag docker_registry in slave/flags.cpp, amd64 images get downloaded
due to which test cases fail on s390x with "Exec format Error"
> 
> 
> Diffs
> -----
> 
>   src/linux/ldcache.cpp 6260496672cd0226a24ea1b34003a48af9678a63 
>   src/slave/flags.cpp 1eccea920338032173be4df6c374ec50dbd2eaf3 
> 
> 
> Diff: https://reviews.apache.org/r/55065/diff/1/
> 
> 
> Testing
> -------
> 
> make check 
> make check GTEST_FILTER="LdcacheTest.Parse"
> 
> 
> Thanks,
> 
> Ayanampudi Varsha
> 
>


Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message