Bug 1634159

Hi Alaa
Can you please work with David and Tim to resolve this.

Hi David,

(In reply to David J. Wilder from comment #0)
> Apply the attached patch to correct build issue on power when MLX support is
> enabled.

Which RHEL and rdma-core versions did you use ?
This issue should not happen with rdma-core v17. It was already fixed (mlx5dv.h will include <sys/types.h>).

Thanks,
Alaa

(In reply to Alaa Hleihel from comment #3)
> Hi David,
> 
> (In reply to David J. Wilder from comment #0)
> > Apply the attached patch to correct build issue on power when MLX support is
> > enabled.
> 
> Which RHEL and rdma-core versions did you use ?
> This issue should not happen with rdma-core v17. It was already fixed
> (mlx5dv.h will include <sys/types.h>).
> 
> Thanks,
> Alaa

Hi Alaa
You are correct,  I upgraded rdma-core to v17.2 and openvswitch built with out my patch and MLX5 enabled.

Thanks
David

Hi Anita,

Following our discussion, we will assign the BZ to Red Hat, to make progress with this request.

Regards,
Noa

Tim, assigning to you as this are "just" config changes now, assuming David will help test the updated packages.
rdma-core on RHEL7 is already at the version mentioned in comment #4.
I don't have much experience with ppc64 but please let me know if I can help anyhow. Thanks

Thanks Timothy,  I will test it.

I am hitting a ovs-vswitchd crash with openvswitch2.10  on p9.
The problem happens when running the PVP test.

gdb -c core
....
Core was generated by `ovs-vswitchd unix:/var/run/openvswitch/db.sock -vconsole:emer -vsyslog:err -vfi'.
Program terminated with signal 11, Segmentation fault.
#0  0x00007fff98210ed0 in mlx5_tx_complete (txq=0x7ff88d88c380)
    at /usr/src/debug/openvswitch-2.10.0/dpdk-17.11/drivers/net/mlx5/mlx5_rxtx.h:481
481					free[blk_n++] = m;

(gdb) list
476		/* Free buffers. */
477		while (elts_free != elts_tail) {
478			m = rte_pktmbuf_prefree_seg((*txq->elts)[elts_free++ & elts_m]);
479			if (likely(m != NULL)) {
480				if (likely(m->pool == pool)) {
481					free[blk_n++] = m;
482				} else {
483					if (likely(pool != NULL))
484						rte_mempool_put_bulk(pool,
485								     (void *)free,

(gdb) print blk_n
$1 = 3241
(gdb) print elts_free
$2 = 916
(gdb) print elts_tail
$3 = 58585
(gdb) print elts_m
$4 = <optimized out>
(gdb) print *txq->elts
Cannot access memory at address 0x7ff88d88c408

I am working with Mellonox on another issue in the same area of the code when running testpmd on the host, it might be related.  I will do more debugging to see if the two problems match.  

Here is some some configuration data.

# rpm -qa | grep openvswitch
openvswitch2.10-2.10.0-10.el7fdb.1.ppc64le
openvswitch2.10-debuginfo-2.10.0-10.el7fdb.1.ppc64le
[root@ltc17u31 /]# rpm -qa | grep rdma-core
rdma-core-15-7.el7_5.ppc64le
rdma-core-devel-15-7.el7_5.ppc64le
[root@ltc17u31 /]# rpm -qa | grep ibacm
ibacm-15-7.el7_5.ppc64le
[root@ltc17u31 /]# rpm -qa | grep libibcm
libibcm-15-7.el7_5.ppc64le
[root@ltc17u31 /]# rpm -qa | grep libibumad
libibumad-15-7.el7_5.ppc64le
[root@ltc17u31 /]# rpm -qa | grep libibverbs
libibverbs-15-7.el7_5.ppc64le
libibverbs-utils-15-7.el7_5.ppc64le
[root@ltc17u31 /]# rpm -qa | grep librdmacm
librdmacm-15-7.el7_5.ppc64le

[root@ltc17u31 /]# uname -a
Linux ltc17u31 4.14.0-49.el7a.ppc64le #1 SMP Wed Mar 14 13:58:40 UTC 2018 ppc64le ppc64le ppc64le GNU/Linux

[root@ltc17u31 /]# ovs-vsctl show
2adc7953-e1bf-4ec7-b39b-42797a141e72
    Bridge "ovs_pvp_br0"
        fail_mode: secure
        Port "dpdk0"
            Interface "dpdk0"
                type: dpdk
                options: {dpdk-devargs="0000:01:00.0", n_rxq="2"}
        Port "ovs_pvp_br0"
            Interface "ovs_pvp_br0"
                type: internal
        Port "vhost0"
            Interface "vhost0"
                type: dpdkvhostuserclient
                options: {n_rxq="2", vhost-server-path="/tmp/vhost-sock0"}
    ovs_version: "2.10.0"

-
dmesg
[ 5636.703686] pmd18[12905]: unhandled signal 11 at 00007fff9ef20000 nip 00007fffa9e60ed0 lr 00007fffa9b1df40 code 1

Created attachment 1497254 [details]
ovs-vswitchd.log, from ovs-vswithd crash.

(In reply to David J. Wilder from comment #10)
> I am hitting a ovs-vswitchd crash with openvswitch2.10  on p9.
> The problem happens when running the PVP test.

Alaa, need your help here.

This is an issue with dpdk and the Mellanox pmd. It is being worked by Mellanox.  RHEL is not currently supporting dpdk on power, so this issue is not blocking OVS.

(In reply to David J. Wilder from comment #13)
> This is an issue with dpdk and the Mellanox pmd. It is being worked by
> Mellanox.  RHEL is not currently supporting dpdk on power, so this issue is
> not blocking OVS.

Thanks David for the update.
BTW, what is the Mellanox support case number ?

Regards,
Alaa

FDB is not released using errata