The ice_read_clock() function was incorrectly using clock_gettime()
with CLOCK_MONOTONIC_RAW to return system time instead of the actual
hardware PHC time. This caused txtime scheduling to use a completely
different time base than the PHC, resulting in all scheduled timestamps
being interpreted as past time by the hardware.
Replace the system clock read with ice_timesync_read_time() which reads
the actual PHC register value, ensuring the application can compute
correct future timestamps relative to the hardware clock.
Fixes: 327fe144ca39 ("net/ice: support clock read")
Cc: [email protected]
Signed-off-by: Soumyadeep Hore <[email protected]>
---
drivers/net/intel/ice/ice_ethdev.c | 16 ++++++++--------
1 file changed, 8 insertions(+), 8 deletions(-)
diff --git a/drivers/net/intel/ice/ice_ethdev.c
b/drivers/net/intel/ice/ice_ethdev.c
index 11b549a452..136780e744 100644
--- a/drivers/net/intel/ice/ice_ethdev.c
+++ b/drivers/net/intel/ice/ice_ethdev.c
@@ -7734,16 +7734,16 @@ ice_timesync_disable(struct rte_eth_dev *dev)
}
static int
-ice_read_clock(__rte_unused struct rte_eth_dev *dev, uint64_t *clock)
+ice_read_clock(struct rte_eth_dev *dev, uint64_t *clock)
{
- struct timespec system_time;
+ struct timespec ts;
+ int ret;
-#ifdef RTE_EXEC_ENV_LINUX
- clock_gettime(CLOCK_MONOTONIC_RAW, &system_time);
-#else
- clock_gettime(CLOCK_MONOTONIC, &system_time);
-#endif
- *clock = system_time.tv_sec * NSEC_PER_SEC + system_time.tv_nsec;
+ ret = ice_timesync_read_time(dev, &ts);
+ if (ret)
+ return ret;
+
+ *clock = (uint64_t)ts.tv_sec * NSEC_PER_SEC + ts.tv_nsec;
return 0;
}
--
2.43.0
