On 3/24/2022 3:09 PM, David Marchand wrote:
A intr_handle is being allocated as a hack to get a (proxy) eventfd from the Linux interrupt implementation. But this handle is never freed.Remove this convoluted hack and create an eventfd in Linux case. Fixes: d61138d4f0e2 ("drivers: remove direct access to interrupt handle") Cc: sta...@dpdk.org Signed-off-by: David Marchand <david.march...@redhat.com> --- drivers/net/failsafe/failsafe_ops.c | 32 ++++++++++------------------- 1 file changed, 11 insertions(+), 21 deletions(-) diff --git a/drivers/net/failsafe/failsafe_ops.c b/drivers/net/failsafe/failsafe_ops.c index 55e21d635c..2c23d0e70a 100644 --- a/drivers/net/failsafe/failsafe_ops.c +++ b/drivers/net/failsafe/failsafe_ops.c @@ -6,6 +6,9 @@ #include <stdbool.h> #include <stdint.h> #include <unistd.h> +#ifdef RTE_EXEC_ENV_LINUX +#include <sys/eventfd.h> +#endif#include <rte_debug.h>#include <rte_atomic.h> @@ -387,28 +390,11 @@ fs_rx_queue_setup(struct rte_eth_dev *dev, const struct rte_eth_rxconf *rx_conf, struct rte_mempool *mb_pool) { - /* - * FIXME: Add a proper interface in rte_eal_interrupts for - * allocating eventfd as an interrupt vector. - * For the time being, fake as if we are using MSIX interrupts, - * this will cause rte_intr_efd_enable to allocate an eventfd for us. - */ - struct rte_intr_handle *intr_handle; struct sub_device *sdev; struct rxq *rxq; uint8_t i; int ret;- intr_handle = rte_intr_instance_alloc(RTE_INTR_INSTANCE_F_PRIVATE);- if (intr_handle == NULL) - return -ENOMEM; - - if (rte_intr_type_set(intr_handle, RTE_INTR_HANDLE_VFIO_MSIX)) - return -rte_errno; - - if (rte_intr_efds_index_set(intr_handle, 0, -1)) - return -rte_errno; - fs_lock(dev, 0); if (rx_conf->rx_deferred_start) { FOREACH_SUBDEV_STATE(sdev, i, dev, DEV_PROBED) { @@ -442,12 +428,16 @@ fs_rx_queue_setup(struct rte_eth_dev *dev, rxq->info.nb_desc = nb_rx_desc; rxq->priv = PRIV(dev); rxq->sdev = PRIV(dev)->subs; - ret = rte_intr_efd_enable(intr_handle, 1); - if (ret < 0) { +#ifdef RTE_EXEC_ENV_LINUX + rxq->event_fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC); + if (rxq->event_fd < 0) { + ERROR("Failed to create an eventfd: %s", strerror(errno)); fs_unlock(dev, 0); - return ret; + return -errno; } - rxq->event_fd = rte_intr_efds_index_get(intr_handle, 0); +#else + rxq->event_fd = -1; +#endif
How this impacts the BSD? I don't know if driver used on BSD but technically it looks supported.
@Gaetan, any objection to the change?
