Red Hat Bugzilla – Bug 1416541
[RHEL-7.4] ibv_*_pingpong manpages missing many options from ibv_*_pingpong --help
Last modified: 2017-08-01 12:57:45 EDT
Description of problem: We can run ibv_rc_pingpong with -t flag to get min/max/average completion clock cycles. However, `man ibv_rc_pingpong` does not say anything about this option. Version-Release number of selected component (if applicable): libibverbs-utils-1.2.1-1.el7.x86_64 How reproducible: always Steps to Reproduce: 1. man ibv_rc_pingpong Actual results: No entry for "-t" flag Expected results: Entry for ibv_rc_pingpong w/ "-t" Additional info: ~$ ibv_rc_pingpong -t 172.31.0.6 local address: LID 0x000e, QPN 0x000081, PSN 0x0441f3, GID :: remote address: LID 0x0022, QPN 0x000075, PSN 0xc1572a, GID :: 8192000 bytes in 0.00 seconds = 15337.23 Mbit/sec 1000 iters in 0.00 seconds = 4.27 usec/iter Max receive completion clock cycles = 2502 Min receive completion clock cycles = 640 Average receive completion clock cycles = 666.791792
There's actually quite a few options missing from the manpage that do show in ibv_rc_pingpong --help. These are: -m, --mtu=<size> path MTU (default 1024) -g, --gid-idx=<gid index> local port gid index -o, --odp use on demand paging -t, --ts get CQE with timestamp ## This also goes for the manpage of ibv_uc_pingpong missing: -m, --mtu=<size> path MTU (default 1024) -g, --gid-idx=<gid index> local port gid index ## The manpage of ibv_ud_pingpong is missing: -g, --gid-idx=<gid index> local port gid index Also, ibv_ud_pingpong manpage has the -l flag: -l, --sl=SL send messages with service level SL (default 0) but it is missing from `ibv_ud_pingpong --help`. ## Plus the manpage ibv_srq_pingpong is missing: -m, --mtu=<size> path MTU (default 1024) -g, --gid-idx=<gid index> local port gid index
Forgot one more: ibv_xsrq_pingpong is missing: -g, --gid-idx=<gid index> local port gid index from its manpage. The reason I had almost forgot about ibv_xsrq_pingpong is because it is not mentioned in the "SEE ALSO" of the other ibv_*_pingpong manpages, even though ibv_xsrq_pingpong points out to the other four. I assume this should be updated in the other four [ibv_rc_pingpong(1), ibv_uc_pingpong(1), ibv_ud_pingpong(1) ibv_srq_pingpong(1)] manpages as well.
Upstream pull request: https://github.com/linux-rdma/rdma-core/pull/62
(In reply to Mike Stowell from comment #3) > Upstream pull request: https://github.com/linux-rdma/rdma-core/pull/62 I just got rdma-core built in Fedora rawhide today, and it'll be making the journey over to RHEL7 shortly as well. I'll be tracking upstream updates needed to sync up with what gets backported kernel-wise, so I suspect we'll have ample opportunity to pick up that fix for free once it's merged (thanks for doing that, btw).
Bug is marked ON_QA but looks like the patch hasn't gotten in to rdma-core-12-2.el7 yet. `man ibv_rc_pingpong` still missing -m/-g/-o/-t options. $ rpm -qf $(which ibv_rc_pingpong) libibverbs-utils-12-2.el7.x86_64 $ man ibv_rc_pingpong <snip> OPTIONS -p, --port=PORT use TCP port PORT for initial synchronization (default 18515) -d, --ib-dev=DEVICE use IB device DEVICE (default first device found) -i, --ib-port=PORT use IB port PORT (default port 1) -s, --size=SIZE ping-pong messages of size SIZE (default 4096) -r, --rx-depth=DEPTH post DEPTH receives at a time (default 1000) -n, --iters=ITERS perform ITERS message exchanges (default 1000) -l, --sl=SL use SL as the service level value of the QP (default 0) -e, --events sleep while waiting for work completion events (default is to poll for completions) </snip>
(In reply to Mike Stowell from comment #6) > Bug is marked ON_QA but looks like the patch hasn't gotten in to > rdma-core-12-2.el7 yet. `man ibv_rc_pingpong` still missing -m/-g/-o/-t > options. D'oh, I thought it was merged upstream prior to the 12 release, hadn't actually checked. I'll have to cherry-pick that patch back into our build then.
Looks good now in libibverbs-utils-12-3.el7.x86_64 provided by rdma-core. Will move to VERIFIED as soon as it hits ON_QA again.
(In reply to Mike Stowell from comment #8) > Looks good now in libibverbs-utils-12-3.el7.x86_64 provided by rdma-core. > Will move to VERIFIED as soon as it hits ON_QA again. Whoops, forgot, I have to move it back to ON_QA, not to MODIFIED, since it's already in the erratum.
Since the problem described in this bug report should be resolved in a recent advisory, it has been closed with a resolution of ERRATA. For information on the advisory, and where to find the updated files, follow the link below. If the solution does not work for you, open a new bug report. https://access.redhat.com/errata/RHEA-2017:2011