)
from lib.py import KsftSkipEx, CmdExitFailure
+# iou-zcrx exits with 42 from setup_zcrx() when the NIC does not advertise
+# QCFG_RX_PAGE_SIZE (or otherwise rejects the requested rx_buf_len).
+SKIP_CODE = 42
+
+
+def _restore_hugepages(count):
+ with open("/proc/sys/vm/nr_hugepages", "w", encoding="utf-8") as f:
+ f.write(str(count))
+
+
+def _mp_clear_wait(cfg, src_queue):
+ """Wait for the io_uring memory provider to clear from the leased
+ physical queue; io_uring tears it down asynchronously after the
+ process holding the ifq exits."""
+ netdevnl = NetdevFamily()
+ deadline = time.time() + 5
+ while time.time() < deadline:
+ queue_info = netdevnl.queue_get(
+ {"ifindex": cfg.ifindex, "id": src_queue, "type": "rx"}
+ )
+ if "io-uring" not in queue_info:
+ return
+ time.sleep(0.1)
+ raise TimeoutError("Timed out waiting for memory provider to clear")
+
def _create_netkit_pair(cfg, rxqueues=2):
if cfg.nk_host_ifname:
cmd(tx_cmd, host=cfg.remote)
+def test_iou_zcrx_large_buf(cfg) -> None:
+ """iou-zcrx with rx_buf_len > page size, going through a netkit-leased
+ queue. Exercises the queue rx-buf-len path via netif_mp_open_rxq()'s
+ lease redirect: the netkit ifindex is opaque to io_uring, but
+ rx_page_size is honoured by the *physical* qops because the lease
+ pointer rewrites the request from netkit onto the leased physical
+ rxq before supported_params/validate_qcfg are consulted.
+ """
+ cfg.require_ipver("6")
+ src_queue, nk_queue = _setup_lease(cfg)
+ defer(_teardown_netkit, cfg)
+ ethnl = EthtoolFamily()
+
+ with open("/proc/sys/vm/nr_hugepages", "r+", encoding="utf-8") as f:
+ nr_hugepages = int(f.read().strip())
+ if nr_hugepages < 64:
+ f.seek(0)
+ f.write("64")
+ defer(_restore_hugepages, nr_hugepages)
+
+ rings = ethnl.rings_get({"header": {"dev-index": cfg.ifindex}})
+ rx_rings = rings["rx"]
+ hds_thresh = rings.get("hds-thresh", 0)
+
+ ethnl.rings_set(
+ {
+ "header": {"dev-index": cfg.ifindex},
+ "tcp-data-split": "enabled",
+ "hds-thresh": 0,
+ "rx": 64,
+ }
+ )
+ defer(
+ ethnl.rings_set,
+ {
+ "header": {"dev-index": cfg.ifindex},
+ "tcp-data-split": "unknown",
+ "hds-thresh": hds_thresh,
+ "rx": rx_rings,
+ },
+ )
+
+ ethtool(f"-X {cfg.ifname} equal {src_queue}")
+ defer(ethtool, f"-X {cfg.ifname} default")
+
+ flow_rule_id = set_flow_rule(cfg, src_queue)
+ defer(ethtool, f"-N {cfg.ifname} delete {flow_rule_id}")
+
+ # -x 2 asks iou-zcrx for rx_buf_len = 2 * page_size (8 KiB on x86_64),
+ # backed by a 2 MiB hugepage area so the chunks are physically
+ # contiguous, which is what zcrx requires for non-default rx_buf_len.
+ rx_cmd = (
+ f"{cfg.bin_local} -s -p {cfg.port} "
+ f"-i {cfg.nk_guest_ifname} -q {nk_queue} -x 2"
+ )
+ tx_cmd = f"{cfg.bin_remote} -c -h {cfg.nk_guest_ipv6} -p {cfg.port} -l 12840"
+
+ # Probe via -d (dry run): exits with SKIP_CODE if the leased physical
+ # qops doesn't advertise QCFG_RX_PAGE_SIZE (e.g. older bnxt FW/HW).
+ probe = cmd(rx_cmd + " -d", fail=False, ns=cfg.netns)
+ if probe.ret == SKIP_CODE:
+ msg = probe.stdout.strip() or "rx_buf_len not supported by leased NIC"
+ raise KsftSkipEx(msg)
+
+ # A successful dry run still registered the zcrx ifq on the leased
+ # physical queue; wait for its async teardown before the real server
+ # binds the same queue.
+ _mp_clear_wait(cfg, src_queue)
+
+ with bkg(rx_cmd, exit_wait=True, ns=cfg.netns):
+ wait_port_listen(cfg.port, proto="tcp", ns=cfg.netns)
+ cmd(tx_cmd, host=cfg.remote)
+
+
def test_attrs(cfg) -> None:
cfg.require_ipver("6")
src_queue, nk_queue = _setup_lease(cfg)
cfg.port = rand_port()
ksft_run(
- [test_iou_zcrx, test_attrs, test_attach_xdp_with_mp, test_destroy],
+ [
+ test_iou_zcrx,
+ test_iou_zcrx_large_buf,
+ test_attrs,
+ test_attach_xdp_with_mp,
+ test_destroy,
+ ],
args=(cfg,),
)
ksft_exit()