Dear Olga and Vincent, thank you for your reply. It seems that the problem has now gone away; both with the -j and without the -j flag. I will try to backtrace the root cause of the problem and report it if I manage to isolate it.
If I may, I would also like to ask your help for a related problem, which I have reported also on the mellanox community forum. If you are available, you may find my question there [1]. I take the occasion to specifically address Olga Shern because I saw that she has already given support on a related topic [2]. Also, related to [2] but possibly un-related from the questions in [1], I am failing to attach a "Perfect Match" filter to my mlx4 NIC, via rte_eth_dev_filter_ctrl(port_id, RTE_ETH_FILTER_FDIR, RTE_ETH_FILTER_ADD, &filter) Is this an expected behavior, namely, mlx4 does not support (yet) this kind of operational mode? Thank you (and anybody willing to help) for your time. Best regards, Harold Demure [1] https://community.mellanox.com/thread/3152 [2] http://dpdk.org/ml/archives/dev/2015-April/016392.html 2016-02-15 9:59 GMT+01:00 Olga Shern <olgas at mellanox.com>: > Hi Harold, > > DPDK 2.2 with MLNX_OFED that you are using and ConnectX-3 support RSS > automatically. > You don?t need to configure anything. > > Looks there is some bug in the compilation, did you compile with make -j, > we saw some issue with it. > Can you please try without -j flag. > > Best Regards, > Olga > > -----Original Message----- > From: users [mailto:users-bounces at dpdk.org] On Behalf Of Harold Demure > Sent: Thursday, February 11, 2016 11:37 PM > To: Vincent Li > Cc: users at dpdk.org > Subject: Re: [dpdk-users] DPDK 2.2 MLX4: problem with number of TX/RX > queues > > Hello Vincent, > Thank you for your reply. I followed the same reasoning as you did, > looking at the source code earlier today. > > Can you send me a link that explains how to change RX for my NIC? > I have been looking around for that, and also performed the steps in the > DPDK official guide for MLX4, but apparently nothing changed (at least, the > error was still there). > > In addition, I have to describe a new strange behavior: even *without > changing anything rss-wise*, but just enabling the increased debugging > verbosity (CONFIG_RTE_LIBRTE_MLX4_DEBUG=y), the error I had suddenly > disappeared. > > I cannot really explain what happened. Can anybody? > > At this point (with the debug prints), now my code blocks because > apparently mlx4 does not support FlowDir, so I have to change a bit my > software, which is built to use it. > > Still, I would love to know what's going on with my rx/tx queues. > > Thank you. > Regards, > Harold > > 2016-02-11 20:10 GMT+01:00 Vincent Li <vincent.mc.li at gmail.com>: > > > On Thu, Feb 11, 2016 at 6:24 AM, Non Voglio > > <harold.demure87 at gmail.com> > > wrote: > > > Hello, > > > I am new to the DPDK world and I am having some troubles in using > > > some code (not written by me). > > > > > > In particular, I bump into this error when invoking the > > rte_eth_dev_configure > > > function: > > > > > > PMD: librte_pmd_mlx4: 0x10a4d2a0: TX queues number update: 0 -> 16 > > > > > > PMD: librte_pmd_mlx4: 0x10a4d2a0: RX queues number update: 0 -> 16 > > > > > > PMD: librte_pmd_mlx4: 0x10a4d2a0: only a single RX queue can be > > configured > > > when hardware doesn't support RSS > > > > > > > it appears you must enable RSS support to use multiple RX queue > > > > if (!priv->hw_rss) { > > ERROR("%p: only a single RX queue can be configured when" > > " hardware doesn't support RSS", > > (void *)dev); > > return EINVAL; > > } > > > > Vincent > > >
