The tap PMD registers each queue file descriptor for Rx interrupts but the feature is not usable through the generic API. This series fixes a crash reachable without any of the interrupt work, then two distinct issues against the original Rx interrupt support, plus a small type cleanup.
Patch 1 fixes a NULL dereference: tap_tx_queue_release() reads dev->data->rx_queues to decide whether the queue fd is still shared, but ethdev releases the Rx queues first and frees that array. It is reached from the reset_queues path of rte_eth_dev_configure() and from rte_eth_dev_internal_reset(), so removing a tap port from a bonding device crashes on current main. This is independent of the rest of the series. Patch 2 adds the missing rx_queue_intr_enable/disable ops. Without them rte_eth_dev_rx_intr_enable() returns -ENOTSUP, so applications that arm queues before sleeping (l3fwd-power and similar) treat tap as having no interrupt support and fall back to polling. Patch 3 fixes a traffic stall specific to interrupt mode: the Rx burst skips reading the queue fd until the SIGIO trigger advances, but the trigger is process wide while the epoll wakeup targets the lcore waiting on that queue, so the burst can run before the trigger has advanced and return 0. The registration is edge triggered, so no further wakeup comes for data that is already queued. The fd is now drained unconditionally in interrupt mode and the SIGIO trigger is not armed on the data queue fds. Patch 4 converts the driver's int-as-boolean fields to bool. Patches 1 to 3 are candidates for stable. v3: - new patch 1 fixing the tap_tx_queue_release() NULL dereference (Stephen), with its own Fixes: and Cc: stable - patch 3: commit message now explains why the trigger can lag behind the epoll wakeup, and that the registration is edge triggered, which is what makes the stall permanent (Stephen); no code change - patch 4: convert the remaining int-as-boolean local and parameters, and use !pmd->flow_init instead of comparing to 0 (Stephen) - rte_eth_tap.h includes <stdbool.h> directly (Stephen) v2: - use bool for the new interrupt-mode fields instead of int (Stephen); both fit existing padding, so neither struct grows - add a patch converting the existing int booleans to bool The coding_style.rst note that discouraged bool in structures no longer matches DPDK practice; it is dropped by a separate patch. Maxime Leroy (4): net/tap: fix crash on Tx queue release after reconfigure net/tap: support Rx queue interrupt enable/disable net/tap: drain queue FD in Rx interrupt mode net/tap: use bool for boolean flags drivers/net/tap/rte_eth_tap.c | 43 ++++++++++++++++++++++++++-------- drivers/net/tap/rte_eth_tap.h | 12 +++++++--- drivers/net/tap/tap_flow.c | 6 ++--- drivers/net/tap/tap_intr.c | 44 +++++++++++++++++++++++++++++++++++ 4 files changed, 89 insertions(+), 16 deletions(-) -- 2.43.0
