On Tue, Aug 11, 2026 at 09:56:35AM +0530, Shaiq Wani wrote:
> The IDPF driver creates vport ethdevs via rte_eth_dev_create() with a
> NULL bus_specific_init callback, bypassing the standard PCI probe path
> that normally copies the PCI device's NUMA node into
> eth_dev->data->numa_node via rte_eth_copy_pci_info().
>
> As a result, rte_eth_dev_socket_id() always returns 0 regardless of
> which NUMA node the device is attached to, causing NUMA-unaware memory
> allocations for mempools and queue descriptors on multi-socket systems.
>
> Set dev->data->numa_node explicitly from the parent PCI device in
> idpf_dev_vport_init() to restore correct NUMA affinity.
>
> Fixes: 5e0f60527e5b ("net/idpf: remove vport req and recv info from adapter")
> Cc: [email protected]
>
> Signed-off-by: Shaiq Wani <[email protected]>
> ---
> drivers/net/intel/idpf/idpf_ethdev.c | 1 +
> 1 file changed, 1 insertion(+)
>
Acked-by: Bruce Richardson <[email protected]>
Applied to dpdk-next-net-intel.
Thanks,
/Bruce
