Hi,

I upgraded from DPDK 17.11 to DPDK 22.11. 

VM host is running Oracle 9 with kernel 5.14.0-162 in VMWARE 7.03 and PCI Passthrough.

Was able to detect the NIC ethernet port with DPDK API (rte_eth_dev_count_avail()), however I see problems when do mmap in our application just after the DPDK initialization.

[root@vstreamOR9-160 bin]# /opt/dpdk/dpdk_nic_bind.py --status

Network devices using DPDK-compatible driver

============================================

0000:0b:00.0 'Ethernet Controller E810-XXV for SFP 159b' drv=igb_uio unused=ice

Network devices using kernel driver

===================================

0000:03:00.0 'VMXNET3 Ethernet Controller 07b0' if=eth0 drv=vmxnet3 unused=igb_uio *Active*

Found errors when application is started with gdb:

EAL: Detected CPU lcores: 16

EAL: Detected NUMA nodes: 1

EAL: Detected static linkage of DPDK

[New Thread 0x7f60c8cb0640 (LWP 3872)]

EAL: Multi-process socket /var/run/dpdk/rte/mp_socket

[New Thread 0x7f60c84af640 (LWP 3873)]

EAL: Selected IOVA mode 'PA'

EAL: rte_mem_virt2phy(): cannot open /proc/self/pagemap: Too many open files

EAL: rte_mem_virt2phy(): cannot open /proc/self/pagemap: Too many open files

EAL: rte_mem_virt2phy(): cannot open /proc/self/pagemap: Too many open files

EAL: rte_mem_virt2phy(): cannot open /proc/self/pagemap: Too many open files

EAL: rte_mem_virt2phy(): cannot open /proc/self/pagemap: Too many open files

EAL: eal_memalloc_alloc_seg_bulk(): couldn't find suitable memseg_list

EAL: FATAL: Cannot init memory

EAL: Cannot init memory

This may be a hint why mmap fails?

Tried the following:

• increase the VM memory from 16 GB to 32 GB,
• increase the amount of hugepages from 2GB to 4GB,
• ulimit -n shows open files = 1024, had to increase to 2048 to eliminate the rte_mem_virt2phy() errors above when start app with gdb.
• free shows 62579808 free and available
• htop shows our application VIRT value is 66 GB.  What a surprise to see this value so high.  With DPDK 17.11 this value is around 5.5 GB.
• pmap -d <app PID> shows there are 4 [ anon ] of 16777216K each, which is what I believe is bumping the VIRT memory.

Was able to get our application to process packets from E810 once I changed the VM total memory to 80GB.

In HTOP the Mem value shows 5.28G/78.5G and VIRT is 69.0 GB.

When the main application executes rte_eal_init() is when the VIRT memory jumps to 66 GB, this is before the application configures the DPDK resources (worker rings, port configuration, etc).

Is there a setting in meson to reduce the DPDK resource in our application?

See the same VIRT memory issue on bare metal with same Oracle 9 and Intel I226 MAC.  I have systems that have only 16 GB of memory and a VM requiring 80 GB is not acceptable for our customers. 

Any help will be appreciated.

Thanks,

Ed