Hello,
On Tue, 17 Nov 2026 at 13:43, Isaac Boukris <[email protected]> wrote:
>
> With Rx interrupts enabled, configuring a port initializes the device,
> and a second time when an offload or RSS is requested. Each
> initialization enables the interrupt eventfds.
>
> EAL now refuses to enable MSI-X eventfds twice, so such a configuration
> fails, and so does configuring a port a second time:
>
> EAL: Interrupt eventfds are already enabled
> virtio_configure_intr(): Fail to create eventfd
> ETHDEV: Port0 dev_configure = -16
>
> Before that EAL change, the previous eventfds were leaked.
>
> Disable the interrupts and release the eventfds before enabling them
> again.
>
> Do it for a VFIO MSI-X handle only, the type whose eventfds EAL
> creates. The other type that gets here is virtio-user's: its eventfds
> are the backend's call fds, set once when the device is created.
> Releasing them would reset their count, which nothing sets again, and
> registering an Rx queue's interrupt would then fail.
>
> To reproduce, with a virtio-net PCI device bound to vfio-pci:
>
> dpdk-l3fwd-power -l 0 -a 0000:0c:00.0 -- -p 1 --interrupt-only \
> --parse-ptype --config="(0,0,0)"
>
> Fixes: 3ba6b689893f ("eal/linux: fix eventfd leaks when enabling interrupts")
> Cc: [email protected]
>
> Signed-off-by: Isaac Boukris <[email protected]>
Thank you for the testing and report.
> ---
> drivers/net/virtio/virtio_ethdev.c | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/drivers/net/virtio/virtio_ethdev.c
> b/drivers/net/virtio/virtio_ethdev.c
> index f2302089ee..37fc896fae 100644
> --- a/drivers/net/virtio/virtio_ethdev.c
> +++ b/drivers/net/virtio/virtio_ethdev.c
> @@ -1455,6 +1455,13 @@ virtio_configure_intr(struct rte_eth_dev *dev)
> return -ENOTSUP;
> }
>
> + /* Reinitialization: release the previous eventfds first. */
> + if (rte_intr_type_get(dev->intr_handle) == RTE_INTR_HANDLE_VFIO_MSIX
> &&
> + rte_intr_dp_is_en(dev->intr_handle)) {
The rte_intr_cap_multiple() guard right before should enough (for the
interrupt type check).
> + virtio_intr_disable(dev);
> + rte_intr_efd_disable(dev->intr_handle);
> + }
> +
Can the number of rxq change at this point?
I don't think it is the case.
If it can't, why do we need to release/set up new eventfds?
The fix could be a simple:
+ if (rte_intr_dp_is_en(dev->intr_handle)) {
> ret = rte_intr_efd_enable(dev->intr_handle, dev->data->nb_rx_queues);
> if (ret < 0) {
> PMD_INIT_LOG(ERR, "Fail to create eventfd");
+ }
We could probably move the rte_intr_vec_list_alloc call under this
check too, but iiuc, this helper does nothing when called multiple
times.
--
David Marchand