> If this is in regard to "rdma_cm segfaults on RoCE with ConnectX-4
> [WAS: Re: rping segfault with 4.9.28 on CentOS 7.3]", I think we have
> narrowed it down to the node GUID being '0' or something along those
> lines. We are still digging into it. We are not getting a kernel
> backtrace when librdmacm segfaults.
The librdmacm associates user space id's with devices based on the node guid. And the check to make this association sees if the node guid from the kernel is non-zero. There may be an assumption further on in the code that a device has been assigned, but one was not, and no error was reported.
This could very well be coming from an error in the kernel not reporting the node_guid correctly, which prevents the librdmacm from making the expected device association.
- Sean
��.n��������+%������w��{.n�����{���fk��ܨ}���Ơz�j:+v�����w����ޙ��&�)ߡ�a����z�ޗ���ݢj��w�f
