On Mon, Jan 16, 2017 at 6:33 PM, Jon Loeliger <j...@netgate.com> wrote:
> Folks,
>
> We are seeing a really strange VPP API message processing problem.
> The scenario goes like this...
>
>
> I've started to isolate it with git bisect, and so far I believe the
> behavior changed between these commits:
>
> $ git log --oneline
> ...
> 553a411 VPP-585: vhost-user interface deletion leaks memory.
> ffc18f5 japi: include tests in jar files
> 738f3f2 API refactoring : dpdk
> 6bbf83a Revert "vppctl: bash completion for vppctl commands"
> 842b9c5 Self-service garbage collection for the API message allocator
> d995c75 vppctl: bash completion for vppctl commands
> 0d3c1cc VPP-279: af_packet via Command-line Arg should no longer be
> supported
> 48c1824 Update CSIT tests 170101 -> 170108
> ...
>
> I've not had a chance to look at them yet, but of those, the most
> suspicious looking one to me is:
> 842b9c5 Self-service garbage collection for the API message allocator
>
> If no one beats me to it, I'll continue bisecting it tomorrow...
>
> Thanks,
> jdl
>
So, our message processing issue was one of the commits
listed above. Specifically, it is this:
738f3f2 API refactoring : dpdk
In that commit, API messages (and their IDs) are made conditionally
present based on the symbol DPDK being greater than 0. That has
yielded an inconsistent build scenario where the include files used
by other components do not have DPDK set, and the internally built
components (vl_api_test) do see it set. That results in API message
IDs that are skewed by 6.
jdl
_______________________________________________
vpp-dev mailing list
vpp-dev@lists.fd.io
https://lists.fd.io/mailman/listinfo/vpp-dev
