e1000e: don't modify SYSTIM registers during SIOCSHWTSTAMP ioctl

PATCH New Jacob Keller 2 participants 3 messages Apr 20, 2016 – Apr 23, 2016

Subsystems (2)

Network drivers Intel Ethernet

lore

Share & tools

lore

Jacob Keller

author

2016-04-20 18:36 UTC #0 lore raw

PATCH v4 1/1 e1000e: don't modify SYSTIM registers during SIOCSHWTSTAMP ioctl

The e1000e_config_hwtstamp function was incorrectly resetting the SYSTIM registers every time the ioctl was being run. If you happened to be running ptp4l and lost the PTP connect (removing cable, or blocking the UDP traffic for example), then ptp4l will eventually perform a restart which involves re-requesting timestamp settings. In e1000e this has the unfortunate and incorrect result of resetting SYSTIME to the kernel time. Since kernel time is usually in UTC, and PTP time is in TAI, this results in the leap second being re-applied.

Fix this by extracting the SYSTIME reset out into its own function, e1000e_ptp_reset, which we call during reset to restore the hardware registers. This function will (a) restart the timecounter based on the new system time, (b) restore the previous PPB setting, and (c) restore the previous hwtstamp settings.

In order to perform (b), I had to modify the adjfreq ptp function pointer to store the old delta each time it is called. This also has the side effect of restoring the correct base timinca register correctly. The driver does not need to explicitly zero the ptp_delta variable since the entire adapter structure comes zero-initialized.

Reported-by Brian Walsh <[email protected]>

Signed-off-by Jacob Keller <[email protected]>

Tested-by Brian Walsh <[email protected]>

---

Notes:
This patch conflicts with Brian's "e1000e fix ptp time reset on newtork interruption", and that patch MUST be removed from the queue before applying this one.

- Changes since v1
* move e1000e_init_ptp prior to reset so that PTP clock is created first
* handle NULL ptp_clock_info gracefully.

- Changes since v2
* fix stupid compilation error
* fix stupid null pointer check

- Changes since v3
* fix issue when using part that doesn't support hw timestamping
* rename e1000e_ptp_reset to more apt e1000e_systim_reset
* refactor code a little to reduce duplicate warning and improve readability

 drivers/net/ethernet/intel/e1000e/e1000.h  |  1 +
 drivers/net/ethernet/intel/e1000e/netdev.c | 68 +++++++++++++++++++++++-------
 drivers/net/ethernet/intel/e1000e/ptp.c    |  2 +
 3 files changed, 55 insertions(+), 16 deletions(-)

drivers/net/ethernet/intel/e1000e/e1000.h +1 -0

index 010e6d61c855..ef96cd11d6d2 100644
@@ -347,6 +347,7 @@ struct e1000_adapter {
 	struct ptp_clock *ptp_clock;
 	struct ptp_clock_info ptp_clock_info;
 	struct pm_qos_request pm_qos_req;
+	s32 ptp_delta;
 
 	u16 eee_advert;
 };

drivers/net/ethernet/intel/e1000e/netdev.c +52 ~~-16~~

104 lines not on this page · raw message

drivers/net/ethernet/intel/e1000e/ptp.c +2 -0

index e2ff3ef75d5d..2e1b17ad52a3 100644
@@ -79,6 +79,8 @@ static int e1000e_phc_adjfreq(struct ptp_clock_info *ptp, s32 delta)
 
 	ew32(TIMINCA, timinca);
 
+	adapter->ptp_delta = delta;
+
 	spin_unlock_irqrestore(&adapter->systim_lock, flags);
 
 	return 0;

Signature · 1 line

2.8.1.493.g7b22612

Keller

2016-04-20 18:39 UTC #1 lore raw

PATCH v4 1/1 e1000e: don't modify SYSTIM registers during SIOCSHWTSTAMP ioctl

Quoted 57 lines · Jacob Keller

The e1000e_config_hwtstamp function was incorrectly resetting the SYSTIM
registers every time the ioctl was being run. If you happened to be running ptp4l and lost the PTP connect (removing cable, or blocking the
UDP traffic for example), then ptp4l will eventually perform a restart
which involves re-requesting timestamp settings. In e1000e this has the
unfortunate and incorrect result of resetting SYSTIME to the kernel time. Since kernel time is usually in UTC, and PTP time is in TAI, this
results in the leap second being re-applied.
Fix this by extracting the SYSTIME reset out into its own function, e1000e_ptp_reset, which we call during reset to restore the hardware registers. This function will (a) restart the timecounter based on the
new system time, (b) restore the previous PPB setting, and (c) restore
the previous hwtstamp settings.
In order to perform (b), I had to modify the adjfreq ptp function pointer to store the old delta each time it is called. This also has the
side effect of restoring the correct base timinca register correctly. The driver does not need to explicitly zero the ptp_delta variable since
the entire adapter structure comes zero-initialized.
Reported-by Brian Walsh <[email protected]>
Signed-off-by Jacob Keller <[email protected]>
Tested-by Brian Walsh <[email protected]>
---
Notes:
????This patch conflicts with Brian's "e1000e fix ptp time reset on newtork
????interruption", and that patch MUST be removed from the queue before applying
????this one.
????
????- Changes since v1
????* move e1000e_init_ptp prior to reset so that PTP clock is created first
????* handle NULL ptp_clock_info gracefully.
????
????- Changes since v2
????* fix stupid compilation error
????* fix stupid null pointer check
????
????- Changes since v3
????* fix issue when using part that doesn't support hw timestamping ????* rename e1000e_ptp_reset to more apt e1000e_systim_reset ????* refactor code a little to reduce duplicate warning and improve readability

Quoted 6 lines

?drivers/net/ethernet/intel/e1000e/e1000.h??|??1 + ?drivers/net/ethernet/intel/e1000e/netdev.c | 68
+++++++++++++++++++++++-------
?drivers/net/ethernet/intel/e1000e/ptp.c????|??2 + ?3 files changed, 55 insertions(+), 16 deletions(-)

For review purposes, here is the diff between v3 and v4

drivers/net/ethernet/intel/e1000e/netdev.c +0 -0

index a272ad9f8029..02e28581dcad 100644
@@ -3874,7 +3874,7 @@ static void e1000_flush_desc_rings(struct e1000_adapter *adapter)

?}
?
?/**
- * e1000e_ptp_reset - reset the timesync registers after a hardware reset
+ * e1000e_systim_reset - reset the timesync registers after a hardware reset ? * @adapter: board private structure
? *
? * When the MAC is reset, all hardware bits for timesync will be reset to the @@ -3882,7 +3882,7 @@ static void e1000_flush_desc_rings(struct e1000_adapter *adapter) ? * Since the clock SYSTIME registers are reset, we will simply restore the ? * cyclecounter to the kernel real clock time.
? **/
-static void e1000e_ptp_reset(struct e1000_adapter *adapter)
+static void e1000e_systim_reset(struct e1000_adapter *adapter)
?{
? struct ptp_clock_info *info = &adapter->ptp_clock_info;
? struct e1000_hw *hw = &adapter->hw;
@@ -3890,23 +3890,26 @@ static void e1000e_ptp_reset(struct e1000_adapter *adapter)
? u32 timinca;
? s32 ret_val;
?
+ if (!(adapter->flags & FLAG_HAS_HW_TIMESTAMP))
+ return;
+
? if (info->adjfreq) {
? /* restore the previous ptp frequency delta */
? ret_val = info->adjfreq(info, adapter->ptp_delta);
- if (ret_val)
- dev_warn(&adapter->pdev->dev,
- ?"Failed to restore TIMINCA clock rate delta: %d\n",
- ?ret_val);
? } else {
+ /* set the default base frequency if no adjustment possible */
? ret_val = e1000e_get_base_timinca(adapter, &timinca);
- if (ret_val)
- dev_warn(&adapter->pdev->dev,
- ?"Failed to set default TIMINCA clock rate: %d\n",
- ?ret_val);
- else
+ if (!ret_val)
? ew32(TIMINCA, timinca);
? }
?
+ if (ret_val) {
+ dev_warn(&adapter->pdev->dev,
+ ?"Failed to restore TIMINCA clock rate delta: %d\n",
+ ?ret_val);
+ return;
+ }
+
? /* reset the systim ns time counter */
? spin_lock_irqsave(&adapter->systim_lock, flags);
? timecounter_init(&adapter->tc, &adapter->cc,
@@ -4097,7 +4100,7 @@ void e1000e_reset(struct e1000_adapter *adapter)
? e1000e_reset_adaptive(hw);
?
? /* restore systim and hwtstamp settings */
- e1000e_ptp_reset(adapter);
+ e1000e_systim_reset(adapter);
?
? /* Set EEE advertisement as appropriate */
? if (adapter->flags2 & FLAG2_HAS_EEE) {

Brown

2016-04-23 01:21 UTC #2 lore raw

PATCH v4 1/1 e1000e: don't modify SYSTIM registers during SIOCSHWTSTAMP ioctl

In reply to Jacob Keller

Quoted 53 lines

From: Intel-wired-lan [mailto:intel-wired-lan-bounces at lists.osuosl.org] On Behalf Of Jacob Keller
Sent: Wednesday, April 20, 2016 11:37 AM
To: Kirsher, Jeffrey T <[email protected]>; Intel Wired LAN <intel- wired-lan at lists.osuosl.org>
Subject: [Intel-wired-lan] [PATCH v4] e1000e: don't modify SYSTIM registers during SIOCSHWTSTAMP ioctl
The e1000e_config_hwtstamp function was incorrectly resetting the SYSTIM registers every time the ioctl was being run. If you happened to be running ptp4l and lost the PTP connect (removing cable, or blocking the UDP traffic for example), then ptp4l will eventually perform a restart which involves re-requesting timestamp settings. In e1000e this has the unfortunate and incorrect result of resetting SYSTIME to the kernel time. Since kernel time is usually in UTC, and PTP time is in TAI, this results in the leap second being re-applied.
Fix this by extracting the SYSTIME reset out into its own function, e1000e_ptp_reset, which we call during reset to restore the hardware registers. This function will (a) restart the timecounter based on the new system time, (b) restore the previous PPB setting, and (c) restore the previous hwtstamp settings.
In order to perform (b), I had to modify the adjfreq ptp function pointer to store the old delta each time it is called. This also has the side effect of restoring the correct base timinca register correctly. The driver does not need to explicitly zero the ptp_delta variable since the entire adapter structure comes zero-initialized.
Reported-by Brian Walsh <[email protected]>
Signed-off-by Jacob Keller <[email protected]>
Tested-by Brian Walsh <[email protected]>
---
Notes:
This patch conflicts with Brian's "e1000e fix ptp time reset on newtork interruption", and that patch MUST be removed from the queue before
applying
this one.
- Changes since v1
* move e1000e_init_ptp prior to reset so that PTP clock is created first
* handle NULL ptp_clock_info gracefully.
- Changes since v2
* fix stupid compilation error
* fix stupid null pointer check
- Changes since v3
* fix issue when using part that doesn't support hw timestamping
* rename e1000e_ptp_reset to more apt e1000e_systim_reset
* refactor code a little to reduce duplicate warning and improve readability

Quoted 5 lines

 drivers/net/ethernet/intel/e1000e/e1000.h  |  1 +
 drivers/net/ethernet/intel/e1000e/netdev.c | 68
+++++++++++++++++++++++-------
 drivers/net/ethernet/intel/e1000e/ptp.c    |  2 +
 3 files changed, 55 insertions(+), 16 deletions(-)

Tested-by Aaron Brown <[email protected]>