[bpf-ci-bot] tc_tunnel/udp_mpls times out on aarch64 due to insufficient connect timeout

[kernel-patches-review-bot]

Summary

The tc_tunnel/udp_mpls subtest fails intermittently on aarch64 with EINPROGRESS (errno 115) because the 1000ms connect() timeout is insufficient for the FOU+MPLS+ipip tunnel path to converge under QEMU emulation. Increasing the timeout to 2000ms and using the TIMEOUT_MS macro consistently fixes the issue.

Failure Details

• Test / Component: tc_tunnel/udp_mpls (subtest temporarily disable tc_bpf test to get CI green #15 of tc_tunnel)
• Frequency: Occasional on aarch64 — observed in at least 1 of every ~5 aarch64 runs; passes consistently on x86_64 and s390x
• Failure mode: Flaky — connect() returns EINPROGRESS (errno 115) due to SO_SNDTIMEO expiry
• Affected architectures: aarch64 (QEMU emulation)
• CI runs observed:
  • https://github.com/kernel-patches/bpf/actions/runs/24334874872 (sole failure: tc_tunnel/udp_mpls)
  • https://github.com/kernel-patches/bpf/actions/runs/22674350732 (referenced in the prior timeout fix)

Root Cause Analysis

The tc_tunnel test suite validates BPF-based tunnel encapsulation/decapsulation. The udp_mpls subtest has the most complex kernel decapsulation path, requiring:

1. FOU (Foo-over-UDP) rx port registration (ip fou add port 6635 ipproto 137)
2. An ipip tunnel in mode any ttl 255
3. MPLS label table setup (65536 entries, route, loopback, testtun0 input enabled)
4. Reverse path filter disabled

After configure_kernel_decapsulation() completes all setup commands and brings testtun0 up, the test immediately tries to establish a TCP connection through the tunnel path. On aarch64 QEMU, the kernel needs more time after interface up for the full data path (BPF encap → FOU/MPLS encap → veth → FOU decap → MPLS routing → server) to become operational.

The client connect() call in connect_client_to_server() (test_tc_tunnel.c:171) has SO_SNDTIMEO set to 1000ms. When the tunnel path isn't ready within this window, connect() returns EINPROGRESS and the test fails.

Contributing factors:

• Commit 2790db208b44 increased the timeout from 500ms to 1000ms but only tested on x86_64, not aarch64 QEMU
• connect_client_to_server() used a hardcoded 1000 instead of the TIMEOUT_MS macro, meaning the timeout wasn't consolidated with the server-side timeout
• udp_mpls is the only subtest combining all three of: FOU, MPLS, and kernel decapsulation (other MPLS variants either skip kernel decap via expect_kern_decap_failure or don't use FOU)

Proposed Fix

The patch (0001-selftests-bpf-Fix-tc_tunnel-udp_mpls-timeout-on-aarc.patch) makes two changes:

1. Increase TIMEOUT_MS from 1000 to 2000 — provides sufficient margin for tunnel path convergence on emulated architectures while still detecting genuine connectivity failures promptly
2. Use TIMEOUT_MS in connect_client_to_server() instead of the hardcoded 1000 — ensures server and client timeouts are consistent and controlled by a single macro

Impact

When this test fails, it is the sole failure in the CI run (1 FAILED out of 6202 subtests), causing the entire aarch64 test_progs job to report failure. This creates noise for unrelated patch submissions and wastes reviewer time investigating false negatives.

References

• Prior timeout fix: 2790db208b44 ("selftests/bpf: Improve tc_tunnel test reliability") — Jiayuan Chen, Mar 2026
• Original test conversion: lore thread
• Lore discussion: https://lore.kernel.org/r/20260312083615.31835-1-jiayuan.chen@linux.dev

[kernel-patches-review-bot]

0001-selftests-bpf-Fix-tc_tunnel-udp_mpls-timeout-on-aarc.patch

From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: BPF CI Bot <bot+bpf-ci@kernel.org>
Date: Sun, 13 Apr 2026 00:00:00 +0000
Subject: [PATCH bpf-next] selftests/bpf: Fix tc_tunnel/udp_mpls timeout on
 aarch64

The tc_tunnel/udp_mpls subtest fails intermittently on aarch64 CI with
EINPROGRESS (errno 115), indicating that the TCP connect() times out
before the FOU+MPLS+ipip tunnel path is fully operational.

Commit 2790db208b44 ("selftests/bpf: Improve tc_tunnel test reliability")
increased the timeout from 500ms to 1000ms to fix similar timeouts on
x86_64 under load. However, on aarch64 (running under QEMU emulation),
the complex tunnel setup in the udp_mpls subtest — which involves FOU
UDP decapsulation, MPLS label routing, rp_filter configuration, and an
ipip tunnel — needs more time to converge after
configure_kernel_decapsulation() returns.

Increase TIMEOUT_MS from 1000ms to 2000ms to accommodate the slower
tunnel path convergence on emulated architectures. Also fix
connect_client_to_server() to use the TIMEOUT_MS macro instead of a
hardcoded value, so the server and client timeouts stay consistent.

Link: https://github.com/kernel-patches/bpf/actions/runs/24334874872
Fixes: 2790db208b44 ("selftests/bpf: Improve tc_tunnel test reliability")
Generated-by: BPF CI Bot ($LLM_MODEL_NAME) <bot+bpf-ci@kernel.org>
Signed-off-by: BPF CI Bot <bot+bpf-ci@kernel.org>
---
 .../selftests/bpf/prog_tests/test_tc_tunnel.c       | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tools/testing/selftests/bpf/prog_tests/test_tc_tunnel.c b/tools/testing/selftests/bpf/prog_tests/test_tc_tunnel.c
index 1aa7c9463980..d2aa51d595ce 100644
--- a/tools/testing/selftests/bpf/prog_tests/test_tc_tunnel.c
+++ b/tools/testing/selftests/bpf/prog_tests/test_tc_tunnel.c
@@ -39,7 +39,7 @@
 #define DEFAULT_TEST_DATA_SIZE	100
 #define GSO_TEST_DATA_SIZE	BUFFER_LEN

-#define TIMEOUT_MS			1000
+#define TIMEOUT_MS			2000
 #define TEST_PORT			8000
 #define UDP_PORT			5555
 #define MPLS_UDP_PORT			6635
@@ -168,7 +168,7 @@ static int check_server_rx_data(struct subtest_cfg *cfg,

 static struct connection *connect_client_to_server(struct subtest_cfg *cfg)
 {
-	struct network_helper_opts opts = {.timeout_ms = 1000};
+	struct network_helper_opts opts = {.timeout_ms = TIMEOUT_MS};
 	int family = cfg->ipproto == 6 ? AF_INET6 : AF_INET;
 	struct connection *conn = NULL;
 	int client_fd, server_fd;
--
2.43.0