[Bug 983] [21.11] net/mlx5: increase in number of required hugepages

[Bug 983] [21.11] net/mlx5: increase in number of required hugepages

[bugzilla]

https://bugs.dpdk.org/show_bug.cgi?id=983

            Bug ID: 983
           Summary: [21.11] net/mlx5: increase in number of required
                    hugepages
           Product: DPDK
           Version: 21.11
          Hardware: x86
                OS: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: Normal
         Component: ethdev
          Assignee: dev@dpdk.org
          Reporter: tudor.cornea@gmail.com
  Target Milestone: future

Greetings,


I'm developing a DPDK application, and recently I've found myself in the need
to upgrade from version 20.11 to version 21.11.

I noticed that when I use a ConnectX-6 NIC in PCI-Passthrough mode it seems to
require me to allocate a few extra hugepages. I have not seen this behavior
when I use the NIC in SR-IOV mode, or with other existing drivers. I've not
seen this in version 20.11.

I think I may have managed to reproduce the same behavior using dpdk-testpmd.
I am using 2 MB hugepages on my local setup.

DPDK 20.11 - Mlx5 Driver

NR_HUGEPAGES=200
mount -p /mnt/hugepages
mount -t hugetlbfs hugetlbfs /mnt/hugepages
sysctl vm.nr_hugepages="${NR_NUGEPAGES}"
./x86_64-native-linuxapp-gcc/app/dpdk-testpmd -l 1-4 --log-level=.*,8

DPDK 21.11 - Mlx5 Driver

NR_HUGEPAGES=220
mount -p /mnt/hugepages
mount -t hugetlbfs hugetlbfs /mnt/hugepages
sysctl vm.nr_hugepages="${NR_NUGEPAGES}"
./x86_64-native-linuxapp-gcc/app/dpdk-testpmd -l 1-4 --log-level=.*,8

It seems that starting with DPDK 21.11, I have to allocate an extra 20
hugepages, otherwise the driver fails in allocating the mbuf pool.

testpmd: create a new mbuf pool <mb_pool_0>: n=171456, size=2176, socket=0
testpmd: preferred mempool ops selected: ring_mp_mc
EAL: Error - exiting with code: 1
  Cause: Creation of mbuf pool for socket 0 failed: Cannot allocate memory

I am trying to keep the number of used hugepages to a minimum, and I was
curious if something changed in the new driver that could determine it to
require more hugepages when managing a PF. I've tried to look at the PMD guide
[1] and then at the code, but I haven't really found anything yet which could
help me explain the increase.

The NIC that I'm using is the following:
00:04.0 Ethernet controller: Mellanox Technologies MT28908 Family [ConnectX-6]

OS Distribution: Ubuntu 20.04

[1] https://doc.dpdk.org/guides-21.11/nics/mlx5.html?highlight=mlx5

-- 
You are receiving this mail because:
You are the assignee for the bug.

[Bug 983] [21.11] net/mlx5: increase in number of required hugepages

[bugzilla]

https://bugs.dpdk.org/show_bug.cgi?id=983

Asaf Penso (asafp@nvidia.com) changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
                 CC|                            |asafp@nvidia.com
         Resolution|---                         |WONTFIX

--- Comment #3 from Asaf Penso (asafp@nvidia.com) ---
Nothing to fix.

-- 
You are receiving this mail because:
You are the assignee for the bug.