GPUDirect RDMA code works in Cuda 5.5 whereas with the modified APIs for cuda 6.0 and above, specified in the Nvidia docs page, the system simply crashes after nvidia_p2p_get_pages() functions returns.
From “dmesg” output,
<4>gpudirect_rdma:: nvidia_p2p_get_pages returned success
<4>gpudirect_rdma:: Number of entries in page table :: 1
<4>gpudirect_rdma:: Page_table Entries : Physical base addr id: 0 value = c0400000
<1>BUG: unable to handle kernel paging request at 0000000000400000
<1>IP: [<0000000000400000>] 0x400000
CentOS Release: 6.4
Graphics card : Tesla K20
Driver Version: 340.29
Cuda Toolkit: 6.5
Please help me to fix the issue.
Thanks.