diff options
author | Jacob Keller <jacob.e.keller@intel.com> | 2019-02-12 22:56:24 +0100 |
---|---|---|
committer | Jeff Kirsher <jeffrey.t.kirsher@intel.com> | 2019-04-17 00:10:38 +0200 |
commit | bf4bf09bdd91a75bb175c172b3f7251a4845f591 (patch) | |
tree | 228509a0e10cbaf4543c723bb9e3f7c12d61ef9a /drivers/net/ethernet/intel/i40e/i40e.h | |
parent | i40e: don't allow changes to HW VLAN stripping on active port VLANs (diff) | |
download | linux-bf4bf09bdd91a75bb175c172b3f7251a4845f591.tar.xz linux-bf4bf09bdd91a75bb175c172b3f7251a4845f591.zip |
i40e: save PTP time before a device reset
In the case where PTP is running on the hardware clock, but the kernel
system time is not being synced, a device reset can mess up the clock
time.
This occurs because we reset the clock time based on the kernel time
every reset. This causes us to potentially completely reset the PTP
time, and can cause unexpected behavior in programs like ptp4l.
Avoid this by saving the PTP time prior to device reset, and then
restoring using that time after the reset.
Directly restoring the PTP time we saved isn't perfect, because time
should have continued running, but the clock will essentially be stopped
during the reset. This is still better than the current solution of
assuming that the PTP HW clock is synced to the CLOCK_REALTIME.
We can do even better, by saving the ktime and calculating
a differential, using ktime_get(). This is based on CLOCK_MONOTONIC, and
allows us to get a fairly precise measure of the time difference between
saving and restoring the time.
Using this, we can update the saved PTP time, and use that as the value
to write to the hardware clock registers. This, of course is not perfect.
However, it does help ensure that the PTP time is restored as close as
feasible to the time it should have been if the reset had not occurred.
During device initialization, continue using the system time as the
source for the creation of the PTP clock, since this is the best known
current time source at driver load.
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Diffstat (limited to 'drivers/net/ethernet/intel/i40e/i40e.h')
-rw-r--r-- | drivers/net/ethernet/intel/i40e/i40e.h | 4 |
1 files changed, 4 insertions, 0 deletions
diff --git a/drivers/net/ethernet/intel/i40e/i40e.h b/drivers/net/ethernet/intel/i40e/i40e.h index fc4cae2fef4f..c4afb852cb57 100644 --- a/drivers/net/ethernet/intel/i40e/i40e.h +++ b/drivers/net/ethernet/intel/i40e/i40e.h @@ -612,6 +612,8 @@ struct i40e_pf { struct sk_buff *ptp_tx_skb; unsigned long ptp_tx_start; struct hwtstamp_config tstamp_config; + struct timespec64 ptp_prev_hw_time; + ktime_t ptp_reset_start; struct mutex tmreg_lock; /* Used to protect the SYSTIME registers. */ u32 ptp_adj_mult; u32 tx_hwtstamp_timeouts; @@ -1108,6 +1110,8 @@ void i40e_ptp_rx_hwtstamp(struct i40e_pf *pf, struct sk_buff *skb, u8 index); void i40e_ptp_set_increment(struct i40e_pf *pf); int i40e_ptp_set_ts_config(struct i40e_pf *pf, struct ifreq *ifr); int i40e_ptp_get_ts_config(struct i40e_pf *pf, struct ifreq *ifr); +void i40e_ptp_save_hw_time(struct i40e_pf *pf); +void i40e_ptp_restore_hw_time(struct i40e_pf *pf); void i40e_ptp_init(struct i40e_pf *pf); void i40e_ptp_stop(struct i40e_pf *pf); int i40e_is_vsi_uplink_mode_veb(struct i40e_vsi *vsi); |