* [PATCH net v2] i40e: fix ptp time increment while link is down
@ 2025-11-30 15:23 Markus Blöchl
2025-11-30 19:27 ` [Intel-wired-lan] " Paul Menzel
` (2 more replies)
0 siblings, 3 replies; 5+ messages in thread
From: Markus Blöchl @ 2025-11-30 15:23 UTC (permalink / raw)
To: Tony Nguyen, Przemek Kitszel, Jacob Keller, Andrew Lunn,
David S. Miller, Eric Dumazet, Jakub Kicinski, Paolo Abeni
Cc: Richard Cochran, Markus Blöchl, intel-wired-lan, netdev,
linux-kernel, Markus Blöchl
When an X710 ethernet port with an active ptp daemon (like the ptp4l and
phc2sys combo) suddenly loses its link and regains it after a while, the
ptp daemon has a hard time to recover synchronization and sometimes
entirely fails to do so.
The issue seems to be related to a wrongly configured increment while the
link is down. This could not be observed with the Intel reference driver.
We identified the fix to appear in Intels official ethernet-linux-i40e
release version 2.17.4.
Include the relevant changes in the kernel version of this driver.
Fixes: beb0dff1251d ("i40e: enable PTP")
Cc: stable@vger.kernel.org
Signed-off-by: Markus Blöchl <markus@blochl.de>
---
Tested with an X710 at 10G link speed and kernel version 6.12.42.
---
Changes in v2:
- Fix kdoc and code formatting
- Rebase onto net tree
- Link to v1: https://lore.kernel.org/r/20251119-i40e_ptp_link_down-v1-1-b351fed254b3@blochl.de
---
drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h | 9 +++
drivers/net/ethernet/intel/i40e/i40e_ptp.c | 69 +++++++++++++++++++++--
drivers/net/ethernet/intel/i40e/i40e_register.h | 9 +++
drivers/net/ethernet/intel/i40e/i40e_type.h | 8 +++
4 files changed, 90 insertions(+), 5 deletions(-)
diff --git a/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h b/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
index cc02a85ad42b..ec176e9569ad 100644
--- a/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
+++ b/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
@@ -1488,6 +1488,15 @@ enum i40e_aq_link_speed {
I40E_LINK_SPEED_25GB = BIT(I40E_LINK_SPEED_25GB_SHIFT),
};
+enum i40e_prt_mac_pcs_link_speed {
+ I40E_PRT_MAC_PCS_LINK_SPEED_UNKNOWN = 0,
+ I40E_PRT_MAC_PCS_LINK_SPEED_100MB,
+ I40E_PRT_MAC_PCS_LINK_SPEED_1GB,
+ I40E_PRT_MAC_PCS_LINK_SPEED_10GB,
+ I40E_PRT_MAC_PCS_LINK_SPEED_40GB,
+ I40E_PRT_MAC_PCS_LINK_SPEED_20GB
+};
+
struct i40e_aqc_module_desc {
u8 oui[3];
u8 reserved1;
diff --git a/drivers/net/ethernet/intel/i40e/i40e_ptp.c b/drivers/net/ethernet/intel/i40e/i40e_ptp.c
index 33535418178b..89abe2f22216 100644
--- a/drivers/net/ethernet/intel/i40e/i40e_ptp.c
+++ b/drivers/net/ethernet/intel/i40e/i40e_ptp.c
@@ -847,6 +847,66 @@ void i40e_ptp_rx_hwtstamp(struct i40e_pf *pf, struct sk_buff *skb, u8 index)
i40e_ptp_convert_to_hwtstamp(skb_hwtstamps(skb), ns);
}
+/**
+ * i40e_ptp_get_link_speed_hw - get the link speed
+ * @pf: Board private structure
+ *
+ * Calculate link speed depending on the link status.
+ *
+ * Return: current link speed.
+ **/
+static enum i40e_aq_link_speed i40e_ptp_get_link_speed_hw(struct i40e_pf *pf)
+{
+ bool link_up = pf->hw.phy.link_info.link_info & I40E_AQ_LINK_UP;
+ enum i40e_aq_link_speed link_speed = I40E_LINK_SPEED_UNKNOWN;
+ struct i40e_hw *hw = &pf->hw;
+
+ if (link_up) {
+ struct i40e_link_status *hw_link_info = &hw->phy.link_info;
+
+ i40e_aq_get_link_info(hw, true, NULL, NULL);
+ link_speed = hw_link_info->link_speed;
+ } else {
+ enum i40e_prt_mac_link_speed prtmac_linksta;
+ u64 prtmac_pcs_linksta;
+
+ prtmac_linksta = (rd32(hw, I40E_PRTMAC_LINKSTA(0)) &
+ I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_MASK) >>
+ I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_SHIFT;
+ if (prtmac_linksta == I40E_PRT_MAC_LINK_SPEED_40GB) {
+ link_speed = I40E_LINK_SPEED_40GB;
+ } else {
+ i40e_aq_debug_read_register(hw,
+ I40E_PRTMAC_PCS_LINK_STATUS1(0),
+ &prtmac_pcs_linksta,
+ NULL);
+
+ prtmac_pcs_linksta = (prtmac_pcs_linksta &
+ I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_MASK) >>
+ I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_SHIFT;
+
+ switch (prtmac_pcs_linksta) {
+ case I40E_PRT_MAC_PCS_LINK_SPEED_100MB:
+ link_speed = I40E_LINK_SPEED_100MB;
+ break;
+ case I40E_PRT_MAC_PCS_LINK_SPEED_1GB:
+ link_speed = I40E_LINK_SPEED_1GB;
+ break;
+ case I40E_PRT_MAC_PCS_LINK_SPEED_10GB:
+ link_speed = I40E_LINK_SPEED_10GB;
+ break;
+ case I40E_PRT_MAC_PCS_LINK_SPEED_20GB:
+ link_speed = I40E_LINK_SPEED_20GB;
+ break;
+ default:
+ link_speed = I40E_LINK_SPEED_UNKNOWN;
+ }
+ }
+ }
+
+ return link_speed;
+}
+
/**
* i40e_ptp_set_increment - Utility function to update clock increment rate
* @pf: Board private structure
@@ -857,16 +917,14 @@ void i40e_ptp_rx_hwtstamp(struct i40e_pf *pf, struct sk_buff *skb, u8 index)
**/
void i40e_ptp_set_increment(struct i40e_pf *pf)
{
- struct i40e_link_status *hw_link_info;
+ enum i40e_aq_link_speed link_speed;
struct i40e_hw *hw = &pf->hw;
u64 incval;
u32 mult;
- hw_link_info = &hw->phy.link_info;
+ link_speed = i40e_ptp_get_link_speed_hw(pf);
- i40e_aq_get_link_info(&pf->hw, true, NULL, NULL);
-
- switch (hw_link_info->link_speed) {
+ switch (link_speed) {
case I40E_LINK_SPEED_10GB:
mult = I40E_PTP_10GB_INCVAL_MULT;
break;
@@ -909,6 +967,7 @@ void i40e_ptp_set_increment(struct i40e_pf *pf)
/* Update the base adjustement value. */
WRITE_ONCE(pf->ptp_adj_mult, mult);
smp_mb(); /* Force the above update. */
+ i40e_ptp_set_1pps_signal_hw(pf);
}
/**
diff --git a/drivers/net/ethernet/intel/i40e/i40e_register.h b/drivers/net/ethernet/intel/i40e/i40e_register.h
index 432afbb64201..c4051dbcc297 100644
--- a/drivers/net/ethernet/intel/i40e/i40e_register.h
+++ b/drivers/net/ethernet/intel/i40e/i40e_register.h
@@ -530,6 +530,15 @@
#define I40E_PRTMAC_HSEC_CTL_TX_PAUSE_REFRESH_TIMER_SHIFT 0
#define I40E_PRTMAC_HSEC_CTL_TX_PAUSE_REFRESH_TIMER_MASK I40E_MASK(0xFFFF, \
I40E_PRTMAC_HSEC_CTL_TX_PAUSE_REFRESH_TIMER_SHIFT)
+/* _i=0...3 */ /* Reset: GLOBR */
+#define I40E_PRTMAC_PCS_LINK_STATUS1(_i) (0x0008C200 + ((_i) * 4))
+#define I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_SHIFT 24
+#define I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_MASK I40E_MASK(0x7, I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_SHIFT)
+#define I40E_PRTMAC_PCS_LINK_STATUS2 0x0008C220
+/* _i=0...3 */ /* Reset: GLOBR */
+#define I40E_PRTMAC_LINKSTA(_i) (0x001E2420 + ((_i) * 4))
+#define I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_SHIFT 27
+#define I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_MASK I40E_MASK(0x7, I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_SHIFT)
#define I40E_GLNVM_FLA 0x000B6108 /* Reset: POR */
#define I40E_GLNVM_FLA_LOCKED_SHIFT 6
#define I40E_GLNVM_FLA_LOCKED_MASK I40E_MASK(0x1, I40E_GLNVM_FLA_LOCKED_SHIFT)
diff --git a/drivers/net/ethernet/intel/i40e/i40e_type.h b/drivers/net/ethernet/intel/i40e/i40e_type.h
index ed8bbdb586da..98c8c5709e5f 100644
--- a/drivers/net/ethernet/intel/i40e/i40e_type.h
+++ b/drivers/net/ethernet/intel/i40e/i40e_type.h
@@ -115,6 +115,14 @@ enum i40e_queue_type {
I40E_QUEUE_TYPE_UNKNOWN
};
+enum i40e_prt_mac_link_speed {
+ I40E_PRT_MAC_LINK_SPEED_100MB = 0,
+ I40E_PRT_MAC_LINK_SPEED_1GB,
+ I40E_PRT_MAC_LINK_SPEED_10GB,
+ I40E_PRT_MAC_LINK_SPEED_40GB,
+ I40E_PRT_MAC_LINK_SPEED_20GB
+};
+
struct i40e_link_status {
enum i40e_aq_phy_type phy_type;
enum i40e_aq_link_speed link_speed;
---
base-commit: e5235eb6cfe02a51256013a78f7b28779a7740d5
change-id: 20251119-i40e_ptp_link_down-47934f9e155d
Best regards,
--
Markus Blöchl <markus@blochl.de>
--
^ permalink raw reply related [flat|nested] 5+ messages in thread
* Re: [Intel-wired-lan] [PATCH net v2] i40e: fix ptp time increment while link is down
2025-11-30 15:23 [PATCH net v2] i40e: fix ptp time increment while link is down Markus Blöchl
@ 2025-11-30 19:27 ` Paul Menzel
2025-12-13 15:39 ` Markus Blöchl
2025-12-02 8:44 ` Loktionov, Aleksandr
2025-12-15 7:14 ` Loktionov, Aleksandr
2 siblings, 1 reply; 5+ messages in thread
From: Paul Menzel @ 2025-11-30 19:27 UTC (permalink / raw)
To: Markus Blöchl
Cc: Tony Nguyen, Przemek Kitszel, Jacob Keller, Andrew Lunn,
David S. Miller, Eric Dumazet, Jakub Kicinski, Paolo Abeni,
Richard Cochran, Markus Blöchl, intel-wired-lan, netdev,
linux-kernel
Dear Markus,
Thank you for the patch.
Am 30.11.25 um 16:23 schrieb Markus Blöchl:
> When an X710 ethernet port with an active ptp daemon (like the ptp4l and
> phc2sys combo) suddenly loses its link and regains it after a while, the
> ptp daemon has a hard time to recover synchronization and sometimes
> entirely fails to do so.
>
> The issue seems to be related to a wrongly configured increment while the
> link is down. This could not be observed with the Intel reference driver.
> We identified the fix to appear in Intels official ethernet-linux-i40e
Intel’s
> release version 2.17.4.
>
> Include the relevant changes in the kernel version of this driver.
>
> Fixes: beb0dff1251d ("i40e: enable PTP")
> Cc: stable@vger.kernel.org
> Signed-off-by: Markus Blöchl <markus@blochl.de>
> ---
> Tested with an X710 at 10G link speed and kernel version 6.12.42.
I’d really appreciate it, if you put the in the commit message above,
and if you provided exact steps for reproducing this in the commit message.
> ---
> Changes in v2:
> - Fix kdoc and code formatting
> - Rebase onto net tree
> - Link to v1: https://lore.kernel.org/r/20251119-i40e_ptp_link_down-v1-1-b351fed254b3@blochl.de
> ---
> drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h | 9 +++
> drivers/net/ethernet/intel/i40e/i40e_ptp.c | 69 +++++++++++++++++++++--
> drivers/net/ethernet/intel/i40e/i40e_register.h | 9 +++
> drivers/net/ethernet/intel/i40e/i40e_type.h | 8 +++
> 4 files changed, 90 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h b/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
> index cc02a85ad42b..ec176e9569ad 100644
> --- a/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
> +++ b/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
> @@ -1488,6 +1488,15 @@ enum i40e_aq_link_speed {
> I40E_LINK_SPEED_25GB = BIT(I40E_LINK_SPEED_25GB_SHIFT),
> };
>
> +enum i40e_prt_mac_pcs_link_speed {
> + I40E_PRT_MAC_PCS_LINK_SPEED_UNKNOWN = 0,
> + I40E_PRT_MAC_PCS_LINK_SPEED_100MB,
> + I40E_PRT_MAC_PCS_LINK_SPEED_1GB,
> + I40E_PRT_MAC_PCS_LINK_SPEED_10GB,
> + I40E_PRT_MAC_PCS_LINK_SPEED_40GB,
> + I40E_PRT_MAC_PCS_LINK_SPEED_20GB
> +};
> +
> struct i40e_aqc_module_desc {
> u8 oui[3];
> u8 reserved1;
> diff --git a/drivers/net/ethernet/intel/i40e/i40e_ptp.c b/drivers/net/ethernet/intel/i40e/i40e_ptp.c
> index 33535418178b..89abe2f22216 100644
> --- a/drivers/net/ethernet/intel/i40e/i40e_ptp.c
> +++ b/drivers/net/ethernet/intel/i40e/i40e_ptp.c
> @@ -847,6 +847,66 @@ void i40e_ptp_rx_hwtstamp(struct i40e_pf *pf, struct sk_buff *skb, u8 index)
> i40e_ptp_convert_to_hwtstamp(skb_hwtstamps(skb), ns);
> }
>
> +/**
> + * i40e_ptp_get_link_speed_hw - get the link speed
> + * @pf: Board private structure
> + *
> + * Calculate link speed depending on the link status.
> + *
> + * Return: current link speed.
> + **/
> +static enum i40e_aq_link_speed i40e_ptp_get_link_speed_hw(struct i40e_pf *pf)
> +{
> + bool link_up = pf->hw.phy.link_info.link_info & I40E_AQ_LINK_UP;
> + enum i40e_aq_link_speed link_speed = I40E_LINK_SPEED_UNKNOWN;
> + struct i40e_hw *hw = &pf->hw;
> +
> + if (link_up) {
> + struct i40e_link_status *hw_link_info = &hw->phy.link_info;
> +
> + i40e_aq_get_link_info(hw, true, NULL, NULL);
> + link_speed = hw_link_info->link_speed;
> + } else {
> + enum i40e_prt_mac_link_speed prtmac_linksta;
> + u64 prtmac_pcs_linksta;
> +
> + prtmac_linksta = (rd32(hw, I40E_PRTMAC_LINKSTA(0)) &
> + I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_MASK) >>
> + I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_SHIFT;
> + if (prtmac_linksta == I40E_PRT_MAC_LINK_SPEED_40GB) {
> + link_speed = I40E_LINK_SPEED_40GB;
> + } else {
> + i40e_aq_debug_read_register(hw,
> + I40E_PRTMAC_PCS_LINK_STATUS1(0),
> + &prtmac_pcs_linksta,
> + NULL);
> +
> + prtmac_pcs_linksta = (prtmac_pcs_linksta &
> + I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_MASK) >>
> + I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_SHIFT;
> +
> + switch (prtmac_pcs_linksta) {
> + case I40E_PRT_MAC_PCS_LINK_SPEED_100MB:
> + link_speed = I40E_LINK_SPEED_100MB;
> + break;
> + case I40E_PRT_MAC_PCS_LINK_SPEED_1GB:
> + link_speed = I40E_LINK_SPEED_1GB;
> + break;
> + case I40E_PRT_MAC_PCS_LINK_SPEED_10GB:
> + link_speed = I40E_LINK_SPEED_10GB;
> + break;
> + case I40E_PRT_MAC_PCS_LINK_SPEED_20GB:
> + link_speed = I40E_LINK_SPEED_20GB;
> + break;
> + default:
> + link_speed = I40E_LINK_SPEED_UNKNOWN;
> + }
> + }
> + }
> +
> + return link_speed;
> +}
> +
> /**
> * i40e_ptp_set_increment - Utility function to update clock increment rate
> * @pf: Board private structure
> @@ -857,16 +917,14 @@ void i40e_ptp_rx_hwtstamp(struct i40e_pf *pf, struct sk_buff *skb, u8 index)
> **/
> void i40e_ptp_set_increment(struct i40e_pf *pf)
> {
> - struct i40e_link_status *hw_link_info;
> + enum i40e_aq_link_speed link_speed;
> struct i40e_hw *hw = &pf->hw;
> u64 incval;
> u32 mult;
>
> - hw_link_info = &hw->phy.link_info;
> + link_speed = i40e_ptp_get_link_speed_hw(pf);
>
> - i40e_aq_get_link_info(&pf->hw, true, NULL, NULL);
> -
> - switch (hw_link_info->link_speed) {
> + switch (link_speed) {
Excuse my ignorance, but could you or the Intel authors explain in the
commit message, why `hw_link_info->link_speed` is incorrect?
> case I40E_LINK_SPEED_10GB:
> mult = I40E_PTP_10GB_INCVAL_MULT;
> break;
> @@ -909,6 +967,7 @@ void i40e_ptp_set_increment(struct i40e_pf *pf)
> /* Update the base adjustement value. */
> WRITE_ONCE(pf->ptp_adj_mult, mult);
> smp_mb(); /* Force the above update. */
> + i40e_ptp_set_1pps_signal_hw(pf);
From the commit message, I would have thought, only this line would be
needed.
> }
>
> /**
> diff --git a/drivers/net/ethernet/intel/i40e/i40e_register.h b/drivers/net/ethernet/intel/i40e/i40e_register.h
> index 432afbb64201..c4051dbcc297 100644
> --- a/drivers/net/ethernet/intel/i40e/i40e_register.h
> +++ b/drivers/net/ethernet/intel/i40e/i40e_register.h
> @@ -530,6 +530,15 @@
> #define I40E_PRTMAC_HSEC_CTL_TX_PAUSE_REFRESH_TIMER_SHIFT 0
> #define I40E_PRTMAC_HSEC_CTL_TX_PAUSE_REFRESH_TIMER_MASK I40E_MASK(0xFFFF, \
> I40E_PRTMAC_HSEC_CTL_TX_PAUSE_REFRESH_TIMER_SHIFT)
> +/* _i=0...3 */ /* Reset: GLOBR */
> +#define I40E_PRTMAC_PCS_LINK_STATUS1(_i) (0x0008C200 + ((_i) * 4))
> +#define I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_SHIFT 24
> +#define I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_MASK I40E_MASK(0x7, I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_SHIFT)
> +#define I40E_PRTMAC_PCS_LINK_STATUS2 0x0008C220
> +/* _i=0...3 */ /* Reset: GLOBR */
> +#define I40E_PRTMAC_LINKSTA(_i) (0x001E2420 + ((_i) * 4))
> +#define I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_SHIFT 27
> +#define I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_MASK I40E_MASK(0x7, I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_SHIFT)
> #define I40E_GLNVM_FLA 0x000B6108 /* Reset: POR */
> #define I40E_GLNVM_FLA_LOCKED_SHIFT 6
> #define I40E_GLNVM_FLA_LOCKED_MASK I40E_MASK(0x1, I40E_GLNVM_FLA_LOCKED_SHIFT)
> diff --git a/drivers/net/ethernet/intel/i40e/i40e_type.h b/drivers/net/ethernet/intel/i40e/i40e_type.h
> index ed8bbdb586da..98c8c5709e5f 100644
> --- a/drivers/net/ethernet/intel/i40e/i40e_type.h
> +++ b/drivers/net/ethernet/intel/i40e/i40e_type.h
> @@ -115,6 +115,14 @@ enum i40e_queue_type {
> I40E_QUEUE_TYPE_UNKNOWN
> };
>
> +enum i40e_prt_mac_link_speed {
> + I40E_PRT_MAC_LINK_SPEED_100MB = 0,
> + I40E_PRT_MAC_LINK_SPEED_1GB,
> + I40E_PRT_MAC_LINK_SPEED_10GB,
> + I40E_PRT_MAC_LINK_SPEED_40GB,
> + I40E_PRT_MAC_LINK_SPEED_20GB
> +};
> +
> struct i40e_link_status {
> enum i40e_aq_phy_type phy_type;
> enum i40e_aq_link_speed link_speed;
Kind regards,
Pul
^ permalink raw reply [flat|nested] 5+ messages in thread
* RE: [Intel-wired-lan] [PATCH net v2] i40e: fix ptp time increment while link is down
2025-11-30 15:23 [PATCH net v2] i40e: fix ptp time increment while link is down Markus Blöchl
2025-11-30 19:27 ` [Intel-wired-lan] " Paul Menzel
@ 2025-12-02 8:44 ` Loktionov, Aleksandr
2025-12-15 7:14 ` Loktionov, Aleksandr
2 siblings, 0 replies; 5+ messages in thread
From: Loktionov, Aleksandr @ 2025-12-02 8:44 UTC (permalink / raw)
To: Markus Blöchl, Nguyen, Anthony L, Kitszel, Przemyslaw,
Keller, Jacob E, Andrew Lunn, David S. Miller, Eric Dumazet,
Jakub Kicinski, Paolo Abeni
Cc: Richard Cochran, Bloechl, Markus,
intel-wired-lan@lists.osuosl.org, netdev@vger.kernel.org,
linux-kernel@vger.kernel.org
> -----Original Message-----
> From: Intel-wired-lan <intel-wired-lan-bounces@osuosl.org> On Behalf
> Of Markus Blöchl
> Sent: Sunday, November 30, 2025 4:24 PM
> To: Nguyen, Anthony L <anthony.l.nguyen@intel.com>; Kitszel,
> Przemyslaw <przemyslaw.kitszel@intel.com>; Keller, Jacob E
> <jacob.e.keller@intel.com>; Andrew Lunn <andrew+netdev@lunn.ch>; David
> S. Miller <davem@davemloft.net>; Eric Dumazet <edumazet@google.com>;
> Jakub Kicinski <kuba@kernel.org>; Paolo Abeni <pabeni@redhat.com>
> Cc: Richard Cochran <richardcochran@gmail.com>; Bloechl, Markus
> <markus.bloechl@ipetronik.com>; intel-wired-lan@lists.osuosl.org;
> netdev@vger.kernel.org; linux-kernel@vger.kernel.org; Markus Blöchl
> <markus@blochl.de>
> Subject: [Intel-wired-lan] [PATCH net v2] i40e: fix ptp time increment
> while link is down
>
> When an X710 ethernet port with an active ptp daemon (like the ptp4l
> and phc2sys combo) suddenly loses its link and regains it after a
> while, the ptp daemon has a hard time to recover synchronization and
> sometimes entirely fails to do so.
>
> The issue seems to be related to a wrongly configured increment while
> the link is down. This could not be observed with the Intel reference
> driver.
> We identified the fix to appear in Intels official ethernet-linux-i40e
> release version 2.17.4.
>
> Include the relevant changes in the kernel version of this driver.
>
> Fixes: beb0dff1251d ("i40e: enable PTP")
> Cc: stable@vger.kernel.org
> Signed-off-by: Markus Blöchl <markus@blochl.de>
> ---
> Tested with an X710 at 10G link speed and kernel version 6.12.42.
> ---
> Changes in v2:
> - Fix kdoc and code formatting
> - Rebase onto net tree
> - Link to v1: https://lore.kernel.org/r/20251119-i40e_ptp_link_down-
> v1-1-b351fed254b3@blochl.de
> ---
> drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h | 9 +++
> drivers/net/ethernet/intel/i40e/i40e_ptp.c | 69
> +++++++++++++++++++++--
> drivers/net/ethernet/intel/i40e/i40e_register.h | 9 +++
> drivers/net/ethernet/intel/i40e/i40e_type.h | 8 +++
> 4 files changed, 90 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
> b/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
> index cc02a85ad42b..ec176e9569ad 100644
> --- a/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
> +++ b/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
> @@ -1488,6 +1488,15 @@ enum i40e_aq_link_speed {
> I40E_LINK_SPEED_25GB = BIT(I40E_LINK_SPEED_25GB_SHIFT),
> };
>
> +enum i40e_prt_mac_pcs_link_speed {
> + I40E_PRT_MAC_PCS_LINK_SPEED_UNKNOWN = 0,
> + I40E_PRT_MAC_PCS_LINK_SPEED_100MB,
> + I40E_PRT_MAC_PCS_LINK_SPEED_1GB,
> + I40E_PRT_MAC_PCS_LINK_SPEED_10GB,
> + I40E_PRT_MAC_PCS_LINK_SPEED_40GB,
> + I40E_PRT_MAC_PCS_LINK_SPEED_20GB
> +};
> +
> struct i40e_aqc_module_desc {
> u8 oui[3];
> u8 reserved1;
> diff --git a/drivers/net/ethernet/intel/i40e/i40e_ptp.c
> b/drivers/net/ethernet/intel/i40e/i40e_ptp.c
> index 33535418178b..89abe2f22216 100644
> --- a/drivers/net/ethernet/intel/i40e/i40e_ptp.c
> +++ b/drivers/net/ethernet/intel/i40e/i40e_ptp.c
> @@ -847,6 +847,66 @@ void i40e_ptp_rx_hwtstamp(struct i40e_pf *pf,
> struct sk_buff *skb, u8 index)
> i40e_ptp_convert_to_hwtstamp(skb_hwtstamps(skb), ns); }
>
> +/**
> + * i40e_ptp_get_link_speed_hw - get the link speed
> + * @pf: Board private structure
> + *
> + * Calculate link speed depending on the link status.
> + *
> + * Return: current link speed.
> + **/
> +static enum i40e_aq_link_speed i40e_ptp_get_link_speed_hw(struct
> +i40e_pf *pf) {
> + bool link_up = pf->hw.phy.link_info.link_info &
> I40E_AQ_LINK_UP;
> + enum i40e_aq_link_speed link_speed = I40E_LINK_SPEED_UNKNOWN;
> + struct i40e_hw *hw = &pf->hw;
> +
> + if (link_up) {
> + struct i40e_link_status *hw_link_info = &hw-
> >phy.link_info;
> +
> + i40e_aq_get_link_info(hw, true, NULL, NULL);
> + link_speed = hw_link_info->link_speed;
> + } else {
> + enum i40e_prt_mac_link_speed prtmac_linksta;
> + u64 prtmac_pcs_linksta;
> +
> + prtmac_linksta = (rd32(hw, I40E_PRTMAC_LINKSTA(0)) &
> + I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_MASK)
> >>
> + I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_SHIFT;
Can you consider to use standard macro?
prtmac_linksta = FIELD_GET(I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_MASK,
rd32(hw, I40E_PRTMAC_LINKSTA(0)))
> + if (prtmac_linksta == I40E_PRT_MAC_LINK_SPEED_40GB) {
> + link_speed = I40E_LINK_SPEED_40GB;
> + } else {
> + i40e_aq_debug_read_register(hw,
...
>
> ---
> base-commit: e5235eb6cfe02a51256013a78f7b28779a7740d5
> change-id: 20251119-i40e_ptp_link_down-47934f9e155d
>
> Best regards,
> --
> Markus Blöchl <markus@blochl.de>
>
>
> --
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [Intel-wired-lan] [PATCH net v2] i40e: fix ptp time increment while link is down
2025-11-30 19:27 ` [Intel-wired-lan] " Paul Menzel
@ 2025-12-13 15:39 ` Markus Blöchl
0 siblings, 0 replies; 5+ messages in thread
From: Markus Blöchl @ 2025-12-13 15:39 UTC (permalink / raw)
To: Paul Menzel
Cc: Tony Nguyen, Przemek Kitszel, Jacob Keller, Andrew Lunn,
David S. Miller, Eric Dumazet, Jakub Kicinski, Paolo Abeni,
Richard Cochran, Markus Blöchl, intel-wired-lan, netdev,
linux-kernel
On Sun, Nov 30, 2025 at 08:27:21PM +0100, Paul Menzel wrote:
> Dear Markus,
>
>
> Thank you for the patch.
>
> Am 30.11.25 um 16:23 schrieb Markus Blöchl:
> > When an X710 ethernet port with an active ptp daemon (like the ptp4l and
> > phc2sys combo) suddenly loses its link and regains it after a while, the
> > ptp daemon has a hard time to recover synchronization and sometimes
> > entirely fails to do so.
> >
> > The issue seems to be related to a wrongly configured increment while the
> > link is down. This could not be observed with the Intel reference driver.
> > We identified the fix to appear in Intels official ethernet-linux-i40e
>
> Intel’s
>
> > release version 2.17.4.
> >
> > Include the relevant changes in the kernel version of this driver.
> >
> > Fixes: beb0dff1251d ("i40e: enable PTP")
> > Cc: stable@vger.kernel.org
> > Signed-off-by: Markus Blöchl <markus@blochl.de>
> > ---
> > Tested with an X710 at 10G link speed and kernel version 6.12.42.
>
> I’d really appreciate it, if you put the in the commit message above, and if
> you provided exact steps for reproducing this in the commit message.
>
Which info are are you missing here?
I assume you dropped the relevant word here.
There is not much to add to the commit message except for the visible
results maybe (and they are a bit verbose):
In our setup we connect one X710 port using a SFP+ DAC cable to any other
device. (Do NOT connect the other end of the cable to the same
controller. That still results in weird behaviour.)
We bring the device up and start phc2sys:
root@DUT:~# phc2sys -c ethsfpp2 -s CLOCK_REALTIME -O 0 -m
phc2sys[2861.416]: ethsfpp2 sys offset 34 s0 freq +24309 delay 468
phc2sys[2862.416]: ethsfpp2 sys offset 28 s2 freq +24303 delay 466
phc2sys[2863.416]: ethsfpp2 sys offset 27 s2 freq +24330 delay 470
phc2sys[2864.416]: ethsfpp2 sys offset -6 s2 freq +24306 delay 469
phc2sys[2865.416]: ethsfpp2 sys offset -11 s2 freq +24299 delay 470
phc2sys[2866.416]: ethsfpp2 sys offset -11 s2 freq +24295 delay 469
phc2sys[2867.416]: ethsfpp2 sys offset -5 s2 freq +24298 delay 471
phc2sys[2868.416]: ethsfpp2 sys offset -6 s2 freq +24296 delay 466
phc2sys[2869.417]: ethsfpp2 sys offset -3 s2 freq +24297 delay 467
-- unplug SFP+ connector here
phc2sys[2870.417]: ethsfpp2 sys offset -910053965 s2 freq -900000000 delay 555
phc2sys[2871.417]: ethsfpp2 sys offset -1815157507 s2 freq -900000000 delay 580
phc2sys[2872.417]: ethsfpp2 sys offset -2720276307 s2 freq -900000000 delay 556
phc2sys[2873.417]: ethsfpp2 sys offset -3625378929 s2 freq -900000000 delay 555
phc2sys[2874.417]: ethsfpp2 sys offset -4530478685 s2 freq -900000000 delay 584
phc2sys[2875.417]: ethsfpp2 sys offset -5435585855 s2 freq -900000000 delay 555
phc2sys[2876.417]: ethsfpp2 sys offset -6340689862 s2 freq -900000000 delay 558
phc2sys[2877.418]: ethsfpp2 sys offset -7245811691 s2 freq -900000000 delay 552
phc2sys[2878.418]: ethsfpp2 sys offset -8150937047 s2 freq -900000000 delay 556
phc2sys[2879.418]: ethsfpp2 sys offset -9056054883 s2 freq -900000000 delay 555
phc2sys[2880.418]: clockcheck: clock jumped backward or running slower than expected!
phc2sys[2880.418]: ethsfpp2 sys offset -9961160837 s0 freq -900000000 delay 577
phc2sys[2881.418]: ethsfpp2 sys offset -10866272666 s2 freq -900000000 delay 555
phc2sys[2882.418]: ethsfpp2 sys offset -11771376439 s2 freq -900000000 delay 552
phc2sys[2883.418]: ethsfpp2 sys offset -12676485774 s2 freq -900000000 delay 555
phc2sys[2884.418]: ethsfpp2 sys offset -13581590974 s2 freq -900000000 delay 554
phc2sys[2885.419]: ethsfpp2 sys offset -14486696148 s2 freq -900000000 delay 555
phc2sys[2886.419]: ethsfpp2 sys offset -15391800951 s2 freq -900000000 delay 552
phc2sys[2887.419]: ethsfpp2 sys offset -16296903380 s2 freq -900000000 delay 555
phc2sys[2888.419]: ethsfpp2 sys offset -17202007779 s2 freq -900000000 delay 553
phc2sys[2889.419]: ethsfpp2 sys offset -18107109951 s2 freq -900000000 delay 554
-- plug SFP+ connector back here
phc2sys[2890.419]: ethsfpp2 sys offset -19012214756 s2 freq -900000000 delay 552
phc2sys[2891.419]: clockcheck: clock jumped backward or running slower than expected!
phc2sys[2891.419]: ethsfpp2 sys offset -19486905092 s0 freq -900000000 delay 466
phc2sys[2892.419]: ethsfpp2 sys offset -19499749450 s2 freq -900000000 delay 468
phc2sys[2893.420]: clockcheck: clock jumped backward or running slower than expected!
phc2sys[2893.420]: ethsfpp2 sys offset -18599585360 s0 freq -900000000 delay 467
phc2sys[2894.420]: ethsfpp2 sys offset -17699424847 s2 freq +24219 delay 465
phc2sys[2895.420]: ethsfpp2 sys offset -17699419372 s2 freq -900000000 delay 470
phc2sys[2896.420]: ethsfpp2 sys offset -16799275857 s2 freq -900000000 delay 468
phc2sys[2897.420]: ethsfpp2 sys offset -15899100892 s2 freq -900000000 delay 470
phc2sys[2898.420]: ethsfpp2 sys offset -14998932608 s2 freq -900000000 delay 469
phc2sys[2899.420]: ethsfpp2 sys offset -14098776398 s2 freq -900000000 delay 463
When we do the same with the patch applied it looks like this:
root@DUT:~# phc2sys -c ethsfpp2 -s CLOCK_REALTIME -O 0 -m
phc2sys[3011.380]: ethsfpp2 sys offset 399464 s0 freq -0 delay 467
phc2sys[3012.380]: ethsfpp2 sys offset 423631 s1 freq +24162 delay 467
phc2sys[3013.380]: ethsfpp2 sys offset 5 s2 freq +24167 delay 469
phc2sys[3014.381]: ethsfpp2 sys offset -2 s2 freq +24161 delay 466
phc2sys[3015.381]: ethsfpp2 sys offset 4 s2 freq +24167 delay 472
phc2sys[3016.381]: ethsfpp2 sys offset -12 s2 freq +24152 delay 463
phc2sys[3017.381]: ethsfpp2 sys offset -7 s2 freq +24153 delay 468
phc2sys[3018.381]: ethsfpp2 sys offset 0 s2 freq +24158 delay 466
phc2sys[3019.381]: ethsfpp2 sys offset -4 s2 freq +24154 delay 466
phc2sys[3020.381]: ethsfpp2 sys offset -2 s2 freq +24155 delay 469
phc2sys[3021.381]: ethsfpp2 sys offset -8 s2 freq +24148 delay 465
phc2sys[3022.381]: ethsfpp2 sys offset -6 s2 freq +24148 delay 464
-- unplug SFP+ connector here
phc2sys[3023.382]: ethsfpp2 sys offset -153373 s2 freq -129221 delay 555
phc2sys[3024.382]: ethsfpp2 sys offset 60 s2 freq -21800 delay 578
phc2sys[3025.382]: ethsfpp2 sys offset 46039 s2 freq +24197 delay 556
phc2sys[3026.382]: ethsfpp2 sys offset 46021 s2 freq +37991 delay 549
phc2sys[3027.382]: ethsfpp2 sys offset 32216 s2 freq +37992 delay 578
phc2sys[3028.382]: ethsfpp2 sys offset 18382 s2 freq +33823 delay 551
phc2sys[3029.382]: ethsfpp2 sys offset 8725 s2 freq +29681 delay 553
phc2sys[3030.382]: ethsfpp2 sys offset 3212 s2 freq +26785 delay 555
phc2sys[3031.383]: ethsfpp2 sys offset 587 s2 freq +25124 delay 556
phc2sys[3032.383]: ethsfpp2 sys offset -383 s2 freq +24330 delay 555
phc2sys[3033.383]: ethsfpp2 sys offset -564 s2 freq +24034 delay 554
phc2sys[3034.383]: ethsfpp2 sys offset -447 s2 freq +23982 delay 557
phc2sys[3035.383]: ethsfpp2 sys offset -262 s2 freq +24033 delay 581
phc2sys[3036.383]: ethsfpp2 sys offset -159 s2 freq +24057 delay 564
phc2sys[3037.383]: ethsfpp2 sys offset -73 s2 freq +24095 delay 564
phc2sys[3038.383]: ethsfpp2 sys offset -30 s2 freq +24116 delay 552
phc2sys[3039.384]: ethsfpp2 sys offset -8 s2 freq +24129 delay 554
phc2sys[3040.384]: ethsfpp2 sys offset -11 s2 freq +24124 delay 552
phc2sys[3041.384]: ethsfpp2 sys offset -2 s2 freq +24130 delay 554
phc2sys[3042.384]: ethsfpp2 sys offset -4 s2 freq +24127 delay 554
phc2sys[3043.384]: ethsfpp2 sys offset 4 s2 freq +24134 delay 557
-- plug SFP+ connector back here
phc2sys[3044.384]: clockcheck: clock jumped forward or running faster than expected!
phc2sys[3044.384]: ethsfpp2 sys offset 5660511605 s0 freq +24134 delay 472
phc2sys[3045.384]: clockcheck: clock jumped forward or running faster than expected!
phc2sys[3045.384]: ethsfpp2 sys offset 5912197029 s0 freq +24134 delay 468
phc2sys[3046.385]: ethsfpp2 sys offset 5912221128 s2 freq +48226 delay 468
phc2sys[3047.385]: ethsfpp2 sys offset 5912197000 s2 freq +900000000 delay 471
phc2sys[3048.385]: ethsfpp2 sys offset 5012065553 s2 freq +900000000 delay 468
phc2sys[3049.385]: ethsfpp2 sys offset 4111949940 s2 freq +900000000 delay 462
phc2sys[3050.385]: ethsfpp2 sys offset 3211840906 s2 freq +900000000 delay 469
phc2sys[3051.385]: ethsfpp2 sys offset 2311737228 s2 freq +900000000 delay 467
phc2sys[3052.385]: ethsfpp2 sys offset 1411628446 s2 freq +900000000 delay 467
phc2sys[3053.385]: ethsfpp2 sys offset 511521784 s2 freq +511570010 delay 471
phc2sys[3054.386]: ethsfpp2 sys offset -99598 s2 freq +153405163 delay 462
phc2sys[3055.386]: ethsfpp2 sys offset -153503011 s2 freq -28130 delay 468
phc2sys[3056.386]: ethsfpp2 sys offset -153451254 s2 freq -46027276 delay 469
phc2sys[3057.386]: ethsfpp2 sys offset -107393800 s2 freq -46005198 delay 468
phc2sys[3058.386]: ethsfpp2 sys offset -61358209 s2 freq -32187747 delay 468
phc2sys[3059.386]: ethsfpp2 sys offset -29141851 s2 freq -18378852 delay 471
phc2sys[3060.386]: ethsfpp2 sys offset -10736285 s2 freq -8715841 delay 465
phc2sys[3061.386]: ethsfpp2 sys offset -1994940 s2 freq -3195382 delay 465
phc2sys[3062.386]: ethsfpp2 sys offset 1225024 s2 freq -573900 delay 469
phc2sys[3063.387]: ethsfpp2 sys offset 1823140 s2 freq +391724 delay 466
phc2sys[3064.387]: ethsfpp2 sys offset 1455492 s2 freq +571018 delay 469
phc2sys[3065.387]: ethsfpp2 sys offset 908511 s2 freq +460684 delay 465
phc2sys[3066.387]: ethsfpp2 sys offset 471888 s2 freq +296614 delay 470
phc2sys[3067.387]: ethsfpp2 sys offset 199351 s2 freq +165644 delay 465
phc2sys[3068.387]: ethsfpp2 sys offset 57808 s2 freq +83906 delay 469
phc2sys[3069.387]: ethsfpp2 sys offset -1991 s2 freq +41450 delay 468
phc2sys[3070.387]: ethsfpp2 sys offset -19323 s2 freq +23520 delay 466
phc2sys[3071.388]: ethsfpp2 sys offset -18729 s2 freq +18317 delay 466
phc2sys[3072.388]: ethsfpp2 sys offset -12925 s2 freq +18503 delay 466
phc2sys[3073.388]: ethsfpp2 sys offset -7311 s2 freq +20239 delay 465
Those large jumps in the time offset still aren't great but they at least can be
recovered in a reasonable amount of time.
For reference some more details on our firmware versions:
# ethtool -i ethsfpp2
driver: i40e
version: 6.12.42-xlog
firmware-version: 9.00 0x8000ceb7 0.0.0
expansion-rom-version:
bus-info: 0000:01:00.1
supports-statistics: yes
supports-test: yes
supports-eeprom-access: yes
supports-register-dump: yes
supports-priv-flags: yes
> > ---
> > Changes in v2:
> > - Fix kdoc and code formatting
> > - Rebase onto net tree
> > - Link to v1: https://lore.kernel.org/r/20251119-i40e_ptp_link_down-v1-1-b351fed254b3@blochl.de
> > ---
> > drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h | 9 +++
> > drivers/net/ethernet/intel/i40e/i40e_ptp.c | 69 +++++++++++++++++++++--
> > drivers/net/ethernet/intel/i40e/i40e_register.h | 9 +++
> > drivers/net/ethernet/intel/i40e/i40e_type.h | 8 +++
> > 4 files changed, 90 insertions(+), 5 deletions(-)
> >
> > diff --git a/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h b/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
> > index cc02a85ad42b..ec176e9569ad 100644
> > --- a/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
> > +++ b/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
> > @@ -1488,6 +1488,15 @@ enum i40e_aq_link_speed {
> > I40E_LINK_SPEED_25GB = BIT(I40E_LINK_SPEED_25GB_SHIFT),
> > };
> > +enum i40e_prt_mac_pcs_link_speed {
> > + I40E_PRT_MAC_PCS_LINK_SPEED_UNKNOWN = 0,
> > + I40E_PRT_MAC_PCS_LINK_SPEED_100MB,
> > + I40E_PRT_MAC_PCS_LINK_SPEED_1GB,
> > + I40E_PRT_MAC_PCS_LINK_SPEED_10GB,
> > + I40E_PRT_MAC_PCS_LINK_SPEED_40GB,
> > + I40E_PRT_MAC_PCS_LINK_SPEED_20GB
> > +};
> > +
> > struct i40e_aqc_module_desc {
> > u8 oui[3];
> > u8 reserved1;
> > diff --git a/drivers/net/ethernet/intel/i40e/i40e_ptp.c b/drivers/net/ethernet/intel/i40e/i40e_ptp.c
> > index 33535418178b..89abe2f22216 100644
> > --- a/drivers/net/ethernet/intel/i40e/i40e_ptp.c
> > +++ b/drivers/net/ethernet/intel/i40e/i40e_ptp.c
> > @@ -847,6 +847,66 @@ void i40e_ptp_rx_hwtstamp(struct i40e_pf *pf, struct sk_buff *skb, u8 index)
> > i40e_ptp_convert_to_hwtstamp(skb_hwtstamps(skb), ns);
> > }
> > +/**
> > + * i40e_ptp_get_link_speed_hw - get the link speed
> > + * @pf: Board private structure
> > + *
> > + * Calculate link speed depending on the link status.
> > + *
> > + * Return: current link speed.
> > + **/
> > +static enum i40e_aq_link_speed i40e_ptp_get_link_speed_hw(struct i40e_pf *pf)
> > +{
> > + bool link_up = pf->hw.phy.link_info.link_info & I40E_AQ_LINK_UP;
> > + enum i40e_aq_link_speed link_speed = I40E_LINK_SPEED_UNKNOWN;
> > + struct i40e_hw *hw = &pf->hw;
> > +
> > + if (link_up) {
> > + struct i40e_link_status *hw_link_info = &hw->phy.link_info;
> > +
> > + i40e_aq_get_link_info(hw, true, NULL, NULL);
> > + link_speed = hw_link_info->link_speed;
> > + } else {
> > + enum i40e_prt_mac_link_speed prtmac_linksta;
> > + u64 prtmac_pcs_linksta;
> > +
> > + prtmac_linksta = (rd32(hw, I40E_PRTMAC_LINKSTA(0)) &
> > + I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_MASK) >>
> > + I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_SHIFT;
> > + if (prtmac_linksta == I40E_PRT_MAC_LINK_SPEED_40GB) {
> > + link_speed = I40E_LINK_SPEED_40GB;
> > + } else {
> > + i40e_aq_debug_read_register(hw,
> > + I40E_PRTMAC_PCS_LINK_STATUS1(0),
> > + &prtmac_pcs_linksta,
> > + NULL);
> > +
> > + prtmac_pcs_linksta = (prtmac_pcs_linksta &
> > + I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_MASK) >>
> > + I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_SHIFT;
> > +
> > + switch (prtmac_pcs_linksta) {
> > + case I40E_PRT_MAC_PCS_LINK_SPEED_100MB:
> > + link_speed = I40E_LINK_SPEED_100MB;
> > + break;
> > + case I40E_PRT_MAC_PCS_LINK_SPEED_1GB:
> > + link_speed = I40E_LINK_SPEED_1GB;
> > + break;
> > + case I40E_PRT_MAC_PCS_LINK_SPEED_10GB:
> > + link_speed = I40E_LINK_SPEED_10GB;
> > + break;
> > + case I40E_PRT_MAC_PCS_LINK_SPEED_20GB:
> > + link_speed = I40E_LINK_SPEED_20GB;
> > + break;
> > + default:
> > + link_speed = I40E_LINK_SPEED_UNKNOWN;
> > + }
> > + }
> > + }
> > +
> > + return link_speed;
> > +}
> > +
> > /**
> > * i40e_ptp_set_increment - Utility function to update clock increment rate
> > * @pf: Board private structure
> > @@ -857,16 +917,14 @@ void i40e_ptp_rx_hwtstamp(struct i40e_pf *pf, struct sk_buff *skb, u8 index)
> > **/
> > void i40e_ptp_set_increment(struct i40e_pf *pf)
> > {
> > - struct i40e_link_status *hw_link_info;
> > + enum i40e_aq_link_speed link_speed;
> > struct i40e_hw *hw = &pf->hw;
> > u64 incval;
> > u32 mult;
> > - hw_link_info = &hw->phy.link_info;
> > + link_speed = i40e_ptp_get_link_speed_hw(pf);
> > - i40e_aq_get_link_info(&pf->hw, true, NULL, NULL);
> > -
> > - switch (hw_link_info->link_speed) {
> > + switch (link_speed) {
>
> Excuse my ignorance, but could you or the Intel authors explain in the
> commit message, why `hw_link_info->link_speed` is incorrect?
>
I admittedly have no idea. I can only assume that the controller might
decide to throttle its internal reference clock independently of the
configured link speed if there is no link anyway.
In any case, what would be the "correct" value for link_speed if there
is no link?
But I'd also appreciate it a lot if someone from Intel could chime in
to shed some light.
> > case I40E_LINK_SPEED_10GB:
> > mult = I40E_PTP_10GB_INCVAL_MULT;
> > break;
> > @@ -909,6 +967,7 @@ void i40e_ptp_set_increment(struct i40e_pf *pf)
> > /* Update the base adjustement value. */
> > WRITE_ONCE(pf->ptp_adj_mult, mult);
> > smp_mb(); /* Force the above update. */
> > + i40e_ptp_set_1pps_signal_hw(pf);
>
> From the commit message, I would have thought, only this line would be
> needed.
This line is actually the least important part, as we don't use any PPS
pin in our setup. To be honest, I am not sure if this was even required.
But it was part of the original patch and doesn't seem to make anything
worse.
>
> > }
> > /**
> > diff --git a/drivers/net/ethernet/intel/i40e/i40e_register.h b/drivers/net/ethernet/intel/i40e/i40e_register.h
> > index 432afbb64201..c4051dbcc297 100644
> > --- a/drivers/net/ethernet/intel/i40e/i40e_register.h
> > +++ b/drivers/net/ethernet/intel/i40e/i40e_register.h
> > @@ -530,6 +530,15 @@
> > #define I40E_PRTMAC_HSEC_CTL_TX_PAUSE_REFRESH_TIMER_SHIFT 0
> > #define I40E_PRTMAC_HSEC_CTL_TX_PAUSE_REFRESH_TIMER_MASK I40E_MASK(0xFFFF, \
> > I40E_PRTMAC_HSEC_CTL_TX_PAUSE_REFRESH_TIMER_SHIFT)
> > +/* _i=0...3 */ /* Reset: GLOBR */
> > +#define I40E_PRTMAC_PCS_LINK_STATUS1(_i) (0x0008C200 + ((_i) * 4))
> > +#define I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_SHIFT 24
> > +#define I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_MASK I40E_MASK(0x7, I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_SHIFT)
> > +#define I40E_PRTMAC_PCS_LINK_STATUS2 0x0008C220
> > +/* _i=0...3 */ /* Reset: GLOBR */
> > +#define I40E_PRTMAC_LINKSTA(_i) (0x001E2420 + ((_i) * 4))
> > +#define I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_SHIFT 27
> > +#define I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_MASK I40E_MASK(0x7, I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_SHIFT)
> > #define I40E_GLNVM_FLA 0x000B6108 /* Reset: POR */
> > #define I40E_GLNVM_FLA_LOCKED_SHIFT 6
> > #define I40E_GLNVM_FLA_LOCKED_MASK I40E_MASK(0x1, I40E_GLNVM_FLA_LOCKED_SHIFT)
> > diff --git a/drivers/net/ethernet/intel/i40e/i40e_type.h b/drivers/net/ethernet/intel/i40e/i40e_type.h
> > index ed8bbdb586da..98c8c5709e5f 100644
> > --- a/drivers/net/ethernet/intel/i40e/i40e_type.h
> > +++ b/drivers/net/ethernet/intel/i40e/i40e_type.h
> > @@ -115,6 +115,14 @@ enum i40e_queue_type {
> > I40E_QUEUE_TYPE_UNKNOWN
> > };
> > +enum i40e_prt_mac_link_speed {
> > + I40E_PRT_MAC_LINK_SPEED_100MB = 0,
> > + I40E_PRT_MAC_LINK_SPEED_1GB,
> > + I40E_PRT_MAC_LINK_SPEED_10GB,
> > + I40E_PRT_MAC_LINK_SPEED_40GB,
> > + I40E_PRT_MAC_LINK_SPEED_20GB
> > +};
> > +
> > struct i40e_link_status {
> > enum i40e_aq_phy_type phy_type;
> > enum i40e_aq_link_speed link_speed;
>
>
> Kind regards,
>
> Pul
Thanks,
Markus
--
^ permalink raw reply [flat|nested] 5+ messages in thread
* RE: [Intel-wired-lan] [PATCH net v2] i40e: fix ptp time increment while link is down
2025-11-30 15:23 [PATCH net v2] i40e: fix ptp time increment while link is down Markus Blöchl
2025-11-30 19:27 ` [Intel-wired-lan] " Paul Menzel
2025-12-02 8:44 ` Loktionov, Aleksandr
@ 2025-12-15 7:14 ` Loktionov, Aleksandr
2 siblings, 0 replies; 5+ messages in thread
From: Loktionov, Aleksandr @ 2025-12-15 7:14 UTC (permalink / raw)
To: Markus Blöchl, Nguyen, Anthony L, Kitszel, Przemyslaw,
Keller, Jacob E, Andrew Lunn, David S. Miller, Eric Dumazet,
Jakub Kicinski, Paolo Abeni
Cc: Richard Cochran, Bloechl, Markus,
intel-wired-lan@lists.osuosl.org, netdev@vger.kernel.org,
linux-kernel@vger.kernel.org
> -----Original Message-----
> From: Intel-wired-lan <intel-wired-lan-bounces@osuosl.org> On Behalf
> Of Markus Blöchl
> Sent: Sunday, November 30, 2025 4:24 PM
> To: Nguyen, Anthony L <anthony.l.nguyen@intel.com>; Kitszel,
> Przemyslaw <przemyslaw.kitszel@intel.com>; Keller, Jacob E
> <jacob.e.keller@intel.com>; Andrew Lunn <andrew+netdev@lunn.ch>; David
> S. Miller <davem@davemloft.net>; Eric Dumazet <edumazet@google.com>;
> Jakub Kicinski <kuba@kernel.org>; Paolo Abeni <pabeni@redhat.com>
> Cc: Richard Cochran <richardcochran@gmail.com>; Bloechl, Markus
> <markus.bloechl@ipetronik.com>; intel-wired-lan@lists.osuosl.org;
> netdev@vger.kernel.org; linux-kernel@vger.kernel.org; Markus Blöchl
> <markus@blochl.de>
> Subject: [Intel-wired-lan] [PATCH net v2] i40e: fix ptp time increment
> while link is down
>
> When an X710 ethernet port with an active ptp daemon (like the ptp4l
> and phc2sys combo) suddenly loses its link and regains it after a
> while, the ptp daemon has a hard time to recover synchronization and
> sometimes entirely fails to do so.
>
> The issue seems to be related to a wrongly configured increment while
> the link is down. This could not be observed with the Intel reference
> driver.
> We identified the fix to appear in Intels official ethernet-linux-i40e
> release version 2.17.4.
>
> Include the relevant changes in the kernel version of this driver.
>
> Fixes: beb0dff1251d ("i40e: enable PTP")
> Cc: stable@vger.kernel.org
> Signed-off-by: Markus Blöchl <markus@blochl.de>
> ---
> Tested with an X710 at 10G link speed and kernel version 6.12.42.
> ---
> Changes in v2:
> - Fix kdoc and code formatting
> - Rebase onto net tree
> - Link to v1: https://lore.kernel.org/r/20251119-i40e_ptp_link_down-
> v1-1-b351fed254b3@blochl.de
> ---
> drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h | 9 +++
> drivers/net/ethernet/intel/i40e/i40e_ptp.c | 69
> +++++++++++++++++++++--
> drivers/net/ethernet/intel/i40e/i40e_register.h | 9 +++
> drivers/net/ethernet/intel/i40e/i40e_type.h | 8 +++
> 4 files changed, 90 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
> b/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
> index cc02a85ad42b..ec176e9569ad 100644
> --- a/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
> +++ b/drivers/net/ethernet/intel/i40e/i40e_adminq_cmd.h
> @@ -1488,6 +1488,15 @@ enum i40e_aq_link_speed {
> I40E_LINK_SPEED_25GB = BIT(I40E_LINK_SPEED_25GB_SHIFT),
> };
>
> +enum i40e_prt_mac_pcs_link_speed {
> + I40E_PRT_MAC_PCS_LINK_SPEED_UNKNOWN = 0,
> + I40E_PRT_MAC_PCS_LINK_SPEED_100MB,
> + I40E_PRT_MAC_PCS_LINK_SPEED_1GB,
> + I40E_PRT_MAC_PCS_LINK_SPEED_10GB,
> + I40E_PRT_MAC_PCS_LINK_SPEED_40GB,
> + I40E_PRT_MAC_PCS_LINK_SPEED_20GB
> +};
> +
> struct i40e_aqc_module_desc {
> u8 oui[3];
> u8 reserved1;
> diff --git a/drivers/net/ethernet/intel/i40e/i40e_ptp.c
> b/drivers/net/ethernet/intel/i40e/i40e_ptp.c
> index 33535418178b..89abe2f22216 100644
> --- a/drivers/net/ethernet/intel/i40e/i40e_ptp.c
> +++ b/drivers/net/ethernet/intel/i40e/i40e_ptp.c
> @@ -847,6 +847,66 @@ void i40e_ptp_rx_hwtstamp(struct i40e_pf *pf,
> struct sk_buff *skb, u8 index)
> i40e_ptp_convert_to_hwtstamp(skb_hwtstamps(skb), ns); }
>
> +/**
> + * i40e_ptp_get_link_speed_hw - get the link speed
> + * @pf: Board private structure
> + *
> + * Calculate link speed depending on the link status.
> + *
> + * Return: current link speed.
> + **/
The function uses hardcoded port index 0 in register accesses (I40E_PRTMAC_LINKSTA(0), I40E_PRTMAC_PCS_LINK_STATUS1(0)),
but this limitation is not documented. Additionally, the return value documentation is vague.
> +static enum i40e_aq_link_speed i40e_ptp_get_link_speed_hw(struct
> +i40e_pf *pf) {
> + bool link_up = pf->hw.phy.link_info.link_info &
> I40E_AQ_LINK_UP;
> + enum i40e_aq_link_speed link_speed = I40E_LINK_SPEED_UNKNOWN;
> + struct i40e_hw *hw = &pf->hw;
> +
> + if (link_up) {
> + struct i40e_link_status *hw_link_info = &hw-
> >phy.link_info;
> +
> + i40e_aq_get_link_info(hw, true, NULL, NULL);
> + link_speed = hw_link_info->link_speed;
> + } else {
> + enum i40e_prt_mac_link_speed prtmac_linksta;
I think Linux kernel coding style discourages implicit conversions between enum types and integer types without explicit cast.
Better declare u32 prtmac_linksta;
> + u64 prtmac_pcs_linksta;
> +
> + prtmac_linksta = (rd32(hw, I40E_PRTMAC_LINKSTA(0)) &
> + I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_MASK)
> >>
> + I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_SHIFT;
> + if (prtmac_linksta == I40E_PRT_MAC_LINK_SPEED_40GB) {
> + link_speed = I40E_LINK_SPEED_40GB;
> + } else {
> + i40e_aq_debug_read_register(hw,
> +
> I40E_PRTMAC_PCS_LINK_STATUS1(0),
> + &prtmac_pcs_linksta,
> + NULL);
Both i40e_aq_get_link_info() and i40e_aq_debug_read_register() return error status codes (i40e_status_code / int),
but these return values are completely ignored. If either admin queue command fails,
the function proceeds with potentially stale or uninitialized data.
> +
> + prtmac_pcs_linksta = (prtmac_pcs_linksta &
> +
> I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_MASK) >>
> +
> I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_SHIFT;
> +
> + switch (prtmac_pcs_linksta) {
> + case I40E_PRT_MAC_PCS_LINK_SPEED_100MB:
> + link_speed = I40E_LINK_SPEED_100MB;
> + break;
> + case I40E_PRT_MAC_PCS_LINK_SPEED_1GB:
> + link_speed = I40E_LINK_SPEED_1GB;
> + break;
> + case I40E_PRT_MAC_PCS_LINK_SPEED_10GB:
> + link_speed = I40E_LINK_SPEED_10GB;
> + break;
> + case I40E_PRT_MAC_PCS_LINK_SPEED_20GB:
> + link_speed = I40E_LINK_SPEED_20GB;
> + break;
> + default:
> + link_speed = I40E_LINK_SPEED_UNKNOWN;
> + }
> + }
> + }
> +
> + return link_speed;
> +}
> +
> /**
> * i40e_ptp_set_increment - Utility function to update clock
> increment rate
> * @pf: Board private structure
> @@ -857,16 +917,14 @@ void i40e_ptp_rx_hwtstamp(struct i40e_pf *pf,
> struct sk_buff *skb, u8 index)
> **/
> void i40e_ptp_set_increment(struct i40e_pf *pf) {
> - struct i40e_link_status *hw_link_info;
> + enum i40e_aq_link_speed link_speed;
> struct i40e_hw *hw = &pf->hw;
> u64 incval;
> u32 mult;
>
> - hw_link_info = &hw->phy.link_info;
> + link_speed = i40e_ptp_get_link_speed_hw(pf);
>
> - i40e_aq_get_link_info(&pf->hw, true, NULL, NULL);
> -
> - switch (hw_link_info->link_speed) {
> + switch (link_speed) {
> case I40E_LINK_SPEED_10GB:
> mult = I40E_PTP_10GB_INCVAL_MULT;
> break;
> @@ -909,6 +967,7 @@ void i40e_ptp_set_increment(struct i40e_pf *pf)
> /* Update the base adjustement value. */
> WRITE_ONCE(pf->ptp_adj_mult, mult);
> smp_mb(); /* Force the above update. */
> + i40e_ptp_set_1pps_signal_hw(pf);
> }
>
> /**
> diff --git a/drivers/net/ethernet/intel/i40e/i40e_register.h
> b/drivers/net/ethernet/intel/i40e/i40e_register.h
> index 432afbb64201..c4051dbcc297 100644
> --- a/drivers/net/ethernet/intel/i40e/i40e_register.h
> +++ b/drivers/net/ethernet/intel/i40e/i40e_register.h
> @@ -530,6 +530,15 @@
> #define I40E_PRTMAC_HSEC_CTL_TX_PAUSE_REFRESH_TIMER_SHIFT 0 #define
> I40E_PRTMAC_HSEC_CTL_TX_PAUSE_REFRESH_TIMER_MASK I40E_MASK(0xFFFF, \
> I40E_PRTMAC_HSEC_CTL_TX_PAUSE_REFRESH_TIMER_SHIFT)
> +/* _i=0...3 */ /* Reset: GLOBR */
> +#define I40E_PRTMAC_PCS_LINK_STATUS1(_i) (0x0008C200 + ((_i) * 4))
> +#define I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_SHIFT 24 #define
> +I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_MASK I40E_MASK(0x7,
> +I40E_PRTMAC_PCS_LINK_STATUS1_LINK_SPEED_SHIFT)
> +#define I40E_PRTMAC_PCS_LINK_STATUS2 0x0008C220
> +/* _i=0...3 */ /* Reset: GLOBR */
> +#define I40E_PRTMAC_LINKSTA(_i) (0x001E2420 + ((_i) * 4)) #define
> +I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_SHIFT 27 #define
> +I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_MASK I40E_MASK(0x7,
> +I40E_PRTMAC_LINKSTA_MAC_LINK_SPEED_SHIFT)
> #define I40E_GLNVM_FLA 0x000B6108 /* Reset: POR */ #define
> I40E_GLNVM_FLA_LOCKED_SHIFT 6 #define I40E_GLNVM_FLA_LOCKED_MASK
> I40E_MASK(0x1, I40E_GLNVM_FLA_LOCKED_SHIFT) diff --git
> a/drivers/net/ethernet/intel/i40e/i40e_type.h
> b/drivers/net/ethernet/intel/i40e/i40e_type.h
> index ed8bbdb586da..98c8c5709e5f 100644
> --- a/drivers/net/ethernet/intel/i40e/i40e_type.h
> +++ b/drivers/net/ethernet/intel/i40e/i40e_type.h
> @@ -115,6 +115,14 @@ enum i40e_queue_type {
> I40E_QUEUE_TYPE_UNKNOWN
> };
>
> +enum i40e_prt_mac_link_speed {
> + I40E_PRT_MAC_LINK_SPEED_100MB = 0,
> + I40E_PRT_MAC_LINK_SPEED_1GB,
> + I40E_PRT_MAC_LINK_SPEED_10GB,
> + I40E_PRT_MAC_LINK_SPEED_40GB,
> + I40E_PRT_MAC_LINK_SPEED_20GB
> +};
> +
> struct i40e_link_status {
> enum i40e_aq_phy_type phy_type;
> enum i40e_aq_link_speed link_speed;
>
> ---
> base-commit: e5235eb6cfe02a51256013a78f7b28779a7740d5
> change-id: 20251119-i40e_ptp_link_down-47934f9e155d
>
> Best regards,
> --
> Markus Blöchl <markus@blochl.de>
>
>
> --
With the best regards
Alex
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2025-12-15 7:14 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2025-11-30 15:23 [PATCH net v2] i40e: fix ptp time increment while link is down Markus Blöchl
2025-11-30 19:27 ` [Intel-wired-lan] " Paul Menzel
2025-12-13 15:39 ` Markus Blöchl
2025-12-02 8:44 ` Loktionov, Aleksandr
2025-12-15 7:14 ` Loktionov, Aleksandr
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).