[PATCH v4 net-next 02/15] selftests: xsk: size the TX UMEM in HW_SW_MAX_RING_SIZE
flat view
WARM1d
REVIEWED: 1 (1M)
From: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Date: 2026-10-16 17:43:53
Also in:
bpf
Subsystem:
bpf [general] (safe dynamic programs and tools), bpf [selftests] (test runners & infrastructure), kernel selftest framework, the rest, xdp sockets (af_xdp) · Maintainers:
Alexei Starovoitov, Daniel Borkmann, Andrii Nakryiko, Eduard Zingerman, Kumar Kartikeya Dwivedi, Shuah Khan, Shuah Khan, Linus Torvalds, Magnus Karlsson, Maciej Fijalkowski
Revision v4 of 4 in this series; 1 review trailer.
Revisions (4)
- v1 [diff vs current]
- v2 [diff vs current]
- v3 [diff vs current]
- v4 current
HW_SW_MAX_RING_SIZE gives the RX UMEM 8k frames and 8k rings, then sends batches of 2k and of the maximum HW TX ring size minus 8. This only works in the single-interface mode, where the TX socket uses the RX UMEM, and veth, which cannot change its ring sizes, skips the case. With two interfaces, TX has its own UMEM of DEFAULT_UMEM_BUFFERS (1k) frames and 2k rings: the in-flight limit in __send_pkts() turns negative for a 2k batch, so TX never sends and RX times out, and the second batch would not fit in the TX ring anyway. The next patch removes the single-interface mode, so size the TX UMEM like the RX one. The shared UMEM setup copies the RX UMEM over the TX one and is not affected. With a HW TX ring of more than 8200 descriptors, the second batch would also outgrow the 8k XSK TX ring, and __send_pkts() would spin forever waiting to reserve it, so cap the batch below the XSK ring size. Reviewed-by: Björn Töpel <bjorn@kernel.org> Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com> --- tools/testing/selftests/bpf/prog_tests/test_xsk.c | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-)
diff --git a/tools/testing/selftests/bpf/prog_tests/test_xsk.c b/tools/testing/selftests/bpf/prog_tests/test_xsk.c
index 2a1362874092..05457f77ca52 100644
--- a/tools/testing/selftests/bpf/prog_tests/test_xsk.c
+++ b/tools/testing/selftests/bpf/prog_tests/test_xsk.c@@ -2504,6 +2504,7 @@ int testapp_hw_sw_min_ring_size(struct test_spec *test) int testapp_hw_sw_max_ring_size(struct test_spec *test) { u32 max_descs = XSK_RING_PROD__DEFAULT_NUM_DESCS * 4; + u32 batch_size; int ret; test->set_ring = true;
@@ -2513,6 +2514,9 @@ int testapp_hw_sw_max_ring_size(struct test_spec *test) test->ifobj_rx->xsk->umem->num_frames = max_descs; test->ifobj_rx->xsk->umem->fill_size = max_descs; test->ifobj_rx->xsk->umem->comp_size = max_descs; + test->ifobj_tx->xsk->umem->num_frames = max_descs; + test->ifobj_tx->xsk->umem->fill_size = max_descs; + test->ifobj_tx->xsk->umem->comp_size = max_descs; test->ifobj_tx->xsk->batch_size = XSK_RING_PROD__DEFAULT_NUM_DESCS; test->ifobj_rx->xsk->batch_size = XSK_RING_PROD__DEFAULT_NUM_DESCS;
@@ -2521,10 +2525,12 @@ int testapp_hw_sw_max_ring_size(struct test_spec *test) return ret; /* Set batch_size to 8152 for testing, as the ice HW ignores the 3 lowest bits when - * updating the Rx HW tail register. + * updating the Rx HW tail register. __send_pkts() waits until a whole batch fits + * in the XSK TX ring, so keep it below that ring's size on NICs with larger rings. */ - test->ifobj_tx->xsk->batch_size = test->ifobj_tx->ring.tx_max_pending - 8; - test->ifobj_rx->xsk->batch_size = test->ifobj_tx->ring.tx_max_pending - 8; + batch_size = min(test->ifobj_tx->ring.tx_max_pending, max_descs) - 8; + test->ifobj_tx->xsk->batch_size = batch_size; + test->ifobj_rx->xsk->batch_size = batch_size; if (pkt_stream_replace(test, max_descs, MIN_PKT_SIZE)) { clean_sockets(test, test->ifobj_tx); clean_sockets(test, test->ifobj_rx);
--
2.43.0