On Tue, Jun 11, 2019 at 07:33:25PM -0400, Honggang Li wrote: > The low 64 bits of cxgb3 and cxgb4 devices' GID are zeros. If the > "provider" was set in the option file, ibacm will fail with > segment fault. > > $ sed -i -e 's/# provider ibacmp 0xFE80000000000000/provider ibacmp 0xFE80000000000000/g' /etc/rdma/ibacm_opts.cfg > $ /usr/sbin/ibacm --systemd > Segmentation fault (core dumped) > $ gdb /usr/sbin/ibacm core.ibacm > (gdb) bt > 0 0x00005625a4809217 in acm_assign_provider (port=0x5625a4bc6f28) at /usr/src/debug/rdma-core-25.0-1.el8.x86_64/ibacm/src/acm.c:2285 > 1 acm_port_up (port=0x5625a4bc6f28) at /usr/src/debug/rdma-core-25.0-1.el8.x86_64/ibacm/src/acm.c:2372 > 2 0x00005625a48073d2 in acm_activate_devices () at /usr/src/debug/rdma-core-25.0-1.el8.x86_64/ibacm/src/acm.c:2564 > 3 main (argc=<optimized out>, argv=<optimized out>) at /usr/src/debug/rdma-core-25.0-1.el8.x86_64/ibacm/src/acm.c:3270 > > Note: The rpm was built with tarball generated from upstream repo. The last > commit is aa41a65ec86bdb9c1c86e57885ee588b39558238. > > acm_open_dev function should not open a umad port for iWARP or RoCE devices. > Signed-off-by: Honggang Li <honli@xxxxxxxxxx> > --- > ibacm/src/acm.c | 26 ++++++++++++++++++++++---- > 1 file changed, 22 insertions(+), 4 deletions(-) Applied to for-next, thanks Jason