patrickmacarthur
commented
6 years ago The configuration looks fine, although it looks like a very small number of QPs and descriptors, so it's possible that we're just not allocating enough mbufs in that case for KNI to use.
What PMD driver are you using?
Maybe try adding something like 8192 or 16384 to the second parameter of the rte_pktmbuf_pool_create calls on lines 1009 and 1024 of src/urdmad/main.c and let me know if that makes the KNI out of memory errors go away.
During the start of urdmad daemon I experience the following output: … PMD: Auto Mode 4 PMD: Support Speeds 48 PMD: Auto Link Speed 0 PMD: Auto Link Speed Mask 0 PMD: Forced Link Speed 0 PMD: Port 0 Link Up - speed 10000 Mbps - full-duplex PMD: Set MAC addr USER1: port 0 up mtu=1500 speed 10 Gbps KNI: Out of memory USER1: rte_eth_tx_burst only 0 of 1 packets KNI: Out of memory USER1: rte_eth_tx_burst only 0 of 1 packets … After a few seconds urdma will crash with a segfault. Is it a problem of the virtual address 0x70000?
What are the correct settings for urdma? Mine are the following:
{ "ports": [ { "ipv4_address": "10.2.0.101", "max_qp" : "2", "rx_desc_count" : 16, "tx_desc_count": 16} ], "socket": "/run/urdma.sock", "eal_args": { "log-level": 7, } }