Re: [dpdk-users] Building mlx5 with rdma-core/OFED

[PATRICK KEROULAS <patrick.keroulas@radio-canada.ca>]

Have you checked that:
- the missing symbols like `mlx5dv_dr_action_create_default_miss` are
defined in mlx5dv.h?
- there is no other mlx5dv.h from previous version installed somewhere?

I recommend to keep the package versions provided for your distro and
downgrade dpdk if necessary.


On Mon, Aug 2, 2021 at 4:18 PM Gerry Wan <gerryw@stanford.edu> wrote:
>
> I am running Ubuntu 20.04, and the mlx5dv.h is indeed in `/usr/include/infiniband/`.
>
> The steps I take to build DPDK are
> $ meson --prefix=$DPDK_PATH build
> $ cd build
> $ sudo ninja install  -> fails here
> $ sudo ldconfig
>
> How can I install the correct versions of rdma-core and libibverbs?
>
> On Mon, Aug 2, 2021 at 12:59 PM PATRICK KEROULAS <patrick.keroulas@radio-canada.ca> wrote:
>>
>> I guess you're running Focal or Groovy release so rdma-core and libibverbs
>> are >v28: https://launchpad.net/rdma-core
>> For this version, your missing symbols are defined:
>> https://github.com/linux-rdma/rdma-core/blob/stable-v28/providers/mlx5/mlx5dv.h
>> I expect this header to be installed in `/usr/include/infiniband/`.
>> Please verify.
>>
>> Maybe something wrong with ninja, try meson or make.
>>
>> On Mon, Aug 2, 2021 at 2:35 PM Gerry Wan <gerryw@stanford.edu> wrote:
>> >
>> > How did you install these packages? Are there extra environment variables I need to set?
>> > After running the OFED uninstall script,
>> >
>> > apt-get install rdma-core
>> > apt-get install libibverbs-dev
>> >
>> > both work fine, but ninja install now fails on
>> >
>> > [1/1756] Linking target drivers/librte_common_mlx5.so.21.1.
>> > FAILED: drivers/librte_common_mlx5.so.21.1
>> > cc  -o drivers/librte_common_mlx5.so.21.1 'drivers/a715181@@rte_common_mlx5@sha/meson-generated_.._rte_common_mlx5.pmd.c.o' 'drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_mlx5_devx_cmds.c.o' 'drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_mlx5_common.c.o' 'drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_mlx5_common_mp.c.o' 'drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_mlx5_common_mr.c.o' 'drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_mlx5_malloc.c.o' 'drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_mlx5_common_pci.c.o' 'drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_mlx5_common_devx.c.o' 'drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_nl.c.o' 'drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_common_os.c.o' 'drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_common_verbs.c.o' 'drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_glue.c.o' -Wl,--as-needed -Wl,--no-undefined -Wl,-O1 -shared -fPIC -Wl,--start-group -Wl,-soname,librte_common_mlx5.so.21 -Wl,--no-as-needed -pthread -lm -ldl -lnuma lib/librte_hash.so.21.1 lib/librte_eal.so.21.1 lib/librte_kvargs.so.21.1 lib/librte_telemetry.so.21.1 lib/librte_ring.so.21.1 lib/librte_rcu.so.21.1 lib/librte_pci.so.21.1 drivers/librte_bus_pci.so.21.1 lib/librte_net.so.21.1 lib/librte_mbuf.so.21.1 lib/librte_mempool.so.21.1 -Wl,--version-script=/home/gerryw/dpdk/dpdk-21.02/drivers/common/mlx5/version.map /usr/lib/x86_64-linux-gnu/libmlx5.so /usr/lib/x86_64-linux-gnu/libibverbs.so -Wl,--end-group '-Wl,-rpath,$ORIGIN/../lib:$ORIGIN/' -Wl,-rpath-link,/home/gerryw/dpdk/dpdk-21.02/build/lib -Wl,-rpath-link,/home/gerryw/dpdk/dpdk-21.02/build/drivers
>> > /usr/bin/ld: drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_glue.c.o: in function `mlx5_glue_dv_create_flow_action_aso':
>> > mlx5_glue.c:(.text+0x21f): undefined reference to `mlx5dv_dr_action_create_aso'
>> > /usr/bin/ld: drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_glue.c.o: in function `mlx5_glue_dr_action_create_dest_array':
>> > mlx5_glue.c:(.text+0x23b): undefined reference to `mlx5dv_dr_action_create_dest_array'
>> > /usr/bin/ld: drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_glue.c.o: in function `mlx5_glue_dr_create_flow_action_sampler':
>> > mlx5_glue.c:(.text+0x25b): undefined reference to `mlx5dv_dr_action_create_flow_sampler'
>> > /usr/bin/ld: drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_glue.c.o: in function `mlx5_glue_dv_alloc_pp':
>> > mlx5_glue.c:(.text+0x28b): undefined reference to `mlx5dv_pp_alloc'
>> > /usr/bin/ld: drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_glue.c.o: in function `mlx5_glue_dr_create_flow_action_default_miss':
>> > mlx5_glue.c:(.text+0x4bb): undefined reference to `mlx5dv_dr_action_create_default_miss'
>> > /usr/bin/ld: drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_glue.c.o: in function `mlx5_glue_dv_free_pp':
>> > mlx5_glue.c:(.text+0x277): undefined reference to `mlx5dv_pp_free'
>> > /usr/bin/ld: drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_glue.c.o: in function `mlx5_glue_dr_reclaim_domain_memory':
>> > mlx5_glue.c:(.text+0x2a7): undefined reference to `mlx5dv_dr_domain_set_reclaim_device_memory'
>> > /usr/bin/ld: drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_glue.c.o: in function `mlx5_glue_devx_port_query':
>> > mlx5_glue.c:(.text+0x325): undefined reference to `mlx5dv_query_devx_port'
>> > /usr/bin/ld: drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_glue.c.o: in function `mlx5_glue_dv_create_flow_action_dest_devx_tir':
>> > mlx5_glue.c:(.text+0x545): undefined reference to `mlx5dv_dr_action_create_dest_devx_tir'
>> > /usr/bin/ld: drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_glue.c.o: in function `mlx5_glue_dr_create_flow_action_pop_vlan':
>> > mlx5_glue.c:(.text+0x655): undefined reference to `mlx5dv_dr_action_create_pop_vlan'
>> > /usr/bin/ld: drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_glue.c.o: in function `mlx5_glue_dr_create_flow_action_push_vlan':
>> > mlx5_glue.c:(.text+0x665): undefined reference to `mlx5dv_dr_action_create_push_vlan'
>> > /usr/bin/ld: drivers/a715181@@tmp_rte_common_mlx5@sta/common_mlx5_linux_mlx5_glue.c.o: in function `mlx5_glue_dr_create_flow_action_dest_port':
>> > mlx5_glue.c:(.text+0x685): undefined reference to `mlx5dv_dr_action_create_dest_ib_port'
>> > collect2: error: ld returned 1 exit status
>> > [2/1756] Compiling C object 'drivers/a715181@@tmp_rte_mempool_octeontx@sta/mempool_octeontx_octeontx_fpavf.c.o'.
>> > ninja: build stopped: subcommand failed.
>> >
>> >
>> > On Mon, Aug 2, 2021 at 6:24 AM PATRICK KEROULAS <patrick.keroulas@radio-canada.ca> wrote:
>> >>
>> >>
>> >> Hello Gerry,
>> >> I went through this a while ago and had to properly remove everything related
>> >> to OFED with the uninstaller. Then, install rdma-core and libibverbs-dev which
>> >> includes infiniband/mlx5dv.h IIRC.
>> >> Regards,
>> >> Patrick
>> >>
>> >> On Sun, Aug 1, 2021 at 8:23 PM Gerry Wan <gerryw@stanford.edu> wrote:
>> >>>
>> >>> I recently tried to reinstall MLX OFED 5.2-2.2.0.0 on my bare metal machine
>> >>> Ubuntu 20.04 (previously using the command ./mlnxofedinstall --dpdk
>> >>> --upstream-libs --with-mft --with-kernel-mft). However, it now fails with
>> >>> the message
>> >>>
>> >>> Checking SW Requirements...
>> >>> Removing old packages...
>> >>> Uninstalling the previous version of MLNX_OFED_LINUX
>> >>> Installing new packages Installing mft-4.16.1...
>> >>> Installing ofed-scripts-5.2...
>> >>> Installing mstflint-4.15.0...
>> >>> Installing mlnx-ofed-kernel-utils-5.2...
>> >>> Installing mlnx-ofed-kernel-dkms-5.2...
>> >>> Failed to install mlnx-ofed-kernel-dkms DEB
>> >>> Collecting debug info...
>> >>> See /tmp/MLNX_OFED_LINUX.9221.logs/mlnx-ofed-kernel-dkms.debinstall.log
>> >>>
>> >>> It appears that my kernel version got upgraded to 5.4.0-80-generic which I
>> >>> think may be the issue. I've also tried to install using
>> >>> --add-kernel-support --kernel 5.4.0-80-generic, with no luck.
>> >>>
>> >>> The DPDK website says that rdma-core is now recommended over OFED, but the
>> >>> documentation for how to use it isn't very clear to me. I've tried to
>> >>> install it via
>> >>>
>> >>> apt-get install rdma-core
>> >>>
>> >>> as well as running the quick start script listed here:
>> >>> https://github.com/linux-rdma/rdma-core, but DPDK installation still fails
>> >>> with
>> >>>
>> >>> ../drivers/common/mlx5/linux/mlx5_glue.h:15:10: fatal error:
>> >>> infiniband/mlx5dv.h: No such file or directory
>> >>> 15 | #include <infiniband/mlx5dv.h>
>> >>>
>> >>> Any advice on getting the mlx5 drivers to work again would be helpful!
>>