Skip to contentSkip to conversation

selftests: drv-net: bump the timeout to 15min

PATCH In review Jakub Kicinski 2 participants 4 messages Aug 12 – Aug 17

Subsystems (3)
lore
Share & tools
lore

Jakub Kicinski

author maintainer
#0 lore raw

PATCH 1/1 selftests: drv-net: bump the timeout to 15min

The GRO test takes extra time on NICs which support HW-GRO and LRO (as it also runs the test cases against the NIC implementation). gro.py takes 6min on bnxt and 10min on mlx5 with debug kernel builds. The delta between drivers is to some extent due to their defaults (mlx5 sillily defaults to queue-per-thread, while bnxt follows netif_get_num_default_rss_queues()).

At the end of the day the defaults don't matter much - on larger machines all drivers will create more queues. We should bump the timeout. Let's give ourselves a 50% margin over mlx5's needs and set the timeout to 15min.

---
CC: [email protected]
CC: [email protected]
CC: [email protected]

---
 tools/testing/selftests/drivers/net/settings | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/testing/selftests/drivers/net/settings +1 -1
index eef533824a3c..e2206265f67c 100644
@@ -1 +1 @@
-timeout=360
+timeout=900
Signature · 1 line
2.55.0
#1 lore raw
Quoted 3 lines · Jakub Kicinski

The GRO test takes extra time on NICs which support HW-GRO and LRO (as it also runs the test cases against the NIC implementation). gro.py takes 6min on bnxt and 10min on mlx5 with debug kernel builds.

10m works for xdp.py but it seems that it isn't enough for gro.py :(

    # ok 265 gro.test.lro_ipip_ip_ecn
    not ok 1 selftests: drivers/net: gro.py # exit=130
    # Totals: pass:0 fail:1 xfail:0 xpass:0 skip:0 error:0

NIPA RUNNER TIMEOUT 600 sec (graceful stop)

https://netdev.bots.linux.dev/logs/hwksft/CX7-dbg/results/776322/test-outputs/0-gro-py/stdout

Quoted 25 lines

The delta between drivers is to some extent due to their defaults (mlx5 sillily defaults to queue-per-thread, while bnxt follows netif_get_num_default_rss_queues()).

At the end of the day the defaults don't matter much - on larger machines all drivers will create more queues. We should bump the timeout. Let's give ourselves a 50% margin over mlx5's needs and set the timeout to 15min.

---
CC: [email protected]
CC: [email protected]
CC: [email protected]

---
 tools/testing/selftests/drivers/net/settings | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/testing/selftests/drivers/net/settings +1 -1
index eef533824a3c..e2206265f67c 100644
@@ -1 +1 @@
-timeout=360
+timeout=900
author maintainer
#2 lore raw
Quoted 1 line · On Thu, 13 Aug 2026 11:04:15 +0300 Nimrod Oren wrote:

NIPA RUNNER TIMEOUT 600 sec (graceful stop)

Blah! Well spotted, thanks.
I thought I fixed that but turns out that's not the timeout I thought :/ Hopefully fixed it for good now (it's an internal runner timeout, I set to 20min). Let's see.

author maintainer
#3 lore raw
In reply to Jakub Kicinski
Quoted 11 lines · On Wed, 12 Aug 2026 10:13:02 -0700 Jakub Kicinski wrote:

The GRO test takes extra time on NICs which support HW-GRO and LRO (as it also runs the test cases against the NIC implementation). gro.py takes 6min on bnxt and 10min on mlx5 with debug kernel builds. The delta between drivers is to some extent due to their defaults (mlx5 sillily defaults to queue-per-thread, while bnxt follows netif_get_num_default_rss_queues()).

At the end of the day the defaults don't matter much - on larger machines all drivers will create more queues. We should bump the timeout. Let's give ourselves a 50% margin over mlx5's needs and set the timeout to 15min.

The gve runner takes 20min on this test. It's getting impractical. Let me send a series - this patch plus a patch to break gro up into subtests.

Signature · 1 line
pw-bot: cr