Skip to contentSkip to conversation

Support exposing raw cycle counters in PTP and mlx5

PATCH Applied Tariq Toukan 7 participants 13 messages Aug 12, 2025 – Sep 9, 2025

Subsystems (5)
lore
Share & tools
lore

Tariq Toukan

author maintainer
#0 lore raw

v2 cover Support exposing raw cycle counters in PTP and mlx5

Hi,

This series by Carolina adds support in ptp and usage in mlx5 for exposing the raw free-running cycle counter of PTP hardware clocks.

This is V2. Find previous one here:
https://lore.kernel.org/all/[email protected]/

Find detailed description by Carolina below [1].

Regards,
Tariq

V2:
- Extend the cover letter with more motivation and use cases.

[1]
This patch series introduces support for exposing the raw free-running cycle counter of PTP hardware clocks. When the device is in free-running mode, it emits timestamps as raw cycle values instead of nanoseconds. These values may be passed directly to user space through:

- fwctl: exposes internal device event records that include raw cycle-based timestamps.

- DPDK: retrieves CQEs that contain raw cycle counters, which are passed to user space unmodified.

To address this, the series introduces two new ioctl commands that allow userspace to query the device's raw cycle counter together with host time:

- PTP_SYS_OFFSET_PRECISE_CYCLES

- PTP_SYS_OFFSET_EXTENDED_CYCLES

These commands work like their existing counterparts but return the device timestamp in cycle units instead of real-time nanoseconds. This allows user space to collect (cycle, time) pairs and build a mapping between the device’s free-running clock and host time.

This can also be useful in the XDP fast path: if a driver inserts the raw cycle value into metadata instead of a real-time timestamp, it can avoid the overhead of converting cycles to time in the kernel. Then userspace can resolve the cycle-to-time mapping using this ioctl when needed.

The ioctl enables user space to correlate those with host time, without requiring the PHC to be synchronized, so long as the drift remains stable during collection.

Adds the new PTP ioctls and integrates support in ptp_ioctl():
- ptp: Add ioctl commands to expose raw cycle counter values

Support for exposing raw cycles in mlx5:
- net/mlx5: Extract MTCTR register read logic into helper function
- net/mlx5: Support getcyclesx and getcrosscycles

Carolina Jubran (3):
ptp: Add ioctl commands to expose raw cycle counter values net/mlx5: Extract MTCTR register read logic into helper function net/mlx5: Support getcyclesx and getcrosscycles

 .../ethernet/mellanox/mlx5/core/lib/clock.c   | 113 ++++++++++++++++--
 drivers/ptp/ptp_chardev.c                     |  34 ++++--
 include/uapi/linux/ptp_clock.h                |   4 +
 3 files changed, 130 insertions(+), 21 deletions(-)

base-commit: bc4c0a48bdad7f225740b8e750fdc1da6d85e1eb

Signature · 1 line
2.40.1
author maintainer
#1 lore raw

PATCH v2 1/3 ptp: Add ioctl commands to expose raw cycle counter values

From: Carolina Jubran <[email protected]>

Introduce two new ioctl commands, PTP_SYS_OFFSET_PRECISE_CYCLES and PTP_SYS_OFFSET_EXTENDED_CYCLES, to allow user space to access the raw free-running cycle counter from PTP devices.

These ioctls are variants of the existing PRECISE and EXTENDED offset queries, but instead of returning device time in realtime, they return the raw cycle counter value.

---
 drivers/ptp/ptp_chardev.c      | 34 ++++++++++++++++++++++++++--------
 include/uapi/linux/ptp_clock.h |  4 ++++
 2 files changed, 30 insertions(+), 8 deletions(-)
drivers/ptp/ptp_chardev.c +26 -8
index 4ca5a464a46a..e9719f365aab 100644
@@ -285,17 +285,21 @@ static long ptp_enable_pps(struct ptp_clock *ptp, bool enable)
 		return ops->enable(ops, &req, enable);
 }
 
-static long ptp_sys_offset_precise(struct ptp_clock *ptp, void __user *arg)
+typedef int (*ptp_crosststamp_fn)(struct ptp_clock_info *,
+				  struct system_device_crosststamp *);
+
+static long ptp_sys_offset_precise(struct ptp_clock *ptp, void __user *arg,
+				   ptp_crosststamp_fn crosststamp_fn)
 {
 	struct ptp_sys_offset_precise precise_offset;
 	struct system_device_crosststamp xtstamp;
 	struct timespec64 ts;
 	int err;
 
-	if (!ptp->info->getcrosststamp)
+	if (!crosststamp_fn)
 		return -EOPNOTSUPP;
 
-	err = ptp->info->getcrosststamp(ptp->info, &xtstamp);
+	err = crosststamp_fn(ptp->info, &xtstamp);
 	if (err)
 		return err;
 
@@ -313,12 +317,17 @@ static long ptp_sys_offset_precise(struct ptp_clock *ptp, void __user *arg)
 	return copy_to_user(arg, &precise_offset, sizeof(precise_offset)) ? -EFAULT : 0;
 }
 
-static long ptp_sys_offset_extended(struct ptp_clock *ptp, void __user *arg)
+typedef int (*ptp_gettimex_fn)(struct ptp_clock_info *,
+			       struct timespec64 *,
+			       struct ptp_system_timestamp *);
+
+static long ptp_sys_offset_extended(struct ptp_clock *ptp, void __user *arg,
+				    ptp_gettimex_fn gettimex_fn)
 {
 	struct ptp_sys_offset_extended *extoff __free(kfree) = NULL;
 	struct ptp_system_timestamp sts;
 
-	if (!ptp->info->gettimex64)
+	if (!gettimex_fn)
 		return -EOPNOTSUPP;
 
 	extoff = memdup_user(arg, sizeof(*extoff));
@@ -346,7 +355,7 @@ static long ptp_sys_offset_extended(struct ptp_clock *ptp, void __user *arg)
 		struct timespec64 ts;
 		int err;
 
-		err = ptp->info->gettimex64(ptp->info, &ts, &sts);
+		err = gettimex_fn(ptp->info, &ts, &sts);
 		if (err)
 			return err;
 
@@ -497,11 +506,13 @@ long ptp_ioctl(struct posix_clock_context *pccontext, unsigned int cmd,
 
 	case PTP_SYS_OFFSET_PRECISE:
 	case PTP_SYS_OFFSET_PRECISE2:
-		return ptp_sys_offset_precise(ptp, argptr);
+		return ptp_sys_offset_precise(ptp, argptr,
+					      ptp->info->getcrosststamp);
 
 	case PTP_SYS_OFFSET_EXTENDED:
 	case PTP_SYS_OFFSET_EXTENDED2:
-		return ptp_sys_offset_extended(ptp, argptr);
+		return ptp_sys_offset_extended(ptp, argptr,
+					       ptp->info->gettimex64);
 
 	case PTP_SYS_OFFSET:
 	case PTP_SYS_OFFSET2:
@@ -523,6 +534,13 @@ long ptp_ioctl(struct posix_clock_context *pccontext, unsigned int cmd,
 	case PTP_MASK_EN_SINGLE:
 		return ptp_mask_en_single(pccontext->private_clkdata, argptr);
 
+	case PTP_SYS_OFFSET_PRECISE_CYCLES:
+		return ptp_sys_offset_precise(ptp, argptr,
+					      ptp->info->getcrosscycles);
+
+	case PTP_SYS_OFFSET_EXTENDED_CYCLES:
+		return ptp_sys_offset_extended(ptp, argptr,
+					       ptp->info->getcyclesx64);
 	default:
 		return -ENOTTY;
 	}
include/uapi/linux/ptp_clock.h +4 -0
index 18eefa6d93d6..65f187b5f0d0 100644
@@ -245,6 +245,10 @@ struct ptp_pin_desc {
 	_IOWR(PTP_CLK_MAGIC, 18, struct ptp_sys_offset_extended)
 #define PTP_MASK_CLEAR_ALL  _IO(PTP_CLK_MAGIC, 19)
 #define PTP_MASK_EN_SINGLE  _IOW(PTP_CLK_MAGIC, 20, unsigned int)
+#define PTP_SYS_OFFSET_PRECISE_CYCLES \
+	_IOWR(PTP_CLK_MAGIC, 21, struct ptp_sys_offset_precise)
+#define PTP_SYS_OFFSET_EXTENDED_CYCLES \
+	_IOWR(PTP_CLK_MAGIC, 22, struct ptp_sys_offset_extended)
 
 struct ptp_extts_event {
 	struct ptp_clock_time t; /* Time event occurred. */
Signature · 1 line
2.40.1
maintainer
#2 lore raw

Re: [PATCH net-next V2 1/3] ptp: Add ioctl commands to expose raw cycle counter values

Quoted 6 lines · Tariq Toukan

From: Carolina Jubran <[email protected]>

Introduce two new ioctl commands, PTP_SYS_OFFSET_PRECISE_CYCLES and PTP_SYS_OFFSET_EXTENDED_CYCLES, to allow user space to access the raw free-running cycle counter from PTP devices.

Quoted 7 lines

These ioctls are variants of the existing PRECISE and EXTENDED offset queries, but instead of returning device time in realtime, they return the raw cycle counter value.

Hi Richard,

can we have a formal ack here?

Thanks,

Paolo

maintainer
#3 lore raw

Re: [PATCH net-next V2 1/3] ptp: Add ioctl commands to expose raw cycle counter values

Quoted 11 lines · Paolo Abeni
Quoted 10 lines · Tariq Toukan

From: Carolina Jubran <[email protected]>

Introduce two new ioctl commands, PTP_SYS_OFFSET_PRECISE_CYCLES and PTP_SYS_OFFSET_EXTENDED_CYCLES, to allow user space to access the raw free-running cycle counter from PTP devices.

These ioctls are variants of the existing PRECISE and EXTENDED offset queries, but instead of returning device time in realtime, they return the raw cycle counter value.

Quoted 7 lines
Quoted 3 lines

Hi Richard,

can we have a formal ack here?

Gentle reminder.

Mark

Quoted 5 lines

Thanks,

Paolo

maintainer
#4 lore raw

Re: [PATCH net-next V2 1/3] ptp: Add ioctl commands to expose raw cycle counter values

Quoted 6 lines · Mark Bloch
Quoted 1 line · Paolo Abeni

can we have a formal ack here?

Gentle reminder.

I'm travelling ATM. Please ping again next week.

Thanks,
Richard

#5 lore raw

Re: [PATCH net-next V2 1/3] ptp: Add ioctl commands to expose raw cycle counter values

Quoted 9 lines · Richard Cochran
Quoted 4 lines · Mark Bloch
Quoted 1 line · Paolo Abeni

can we have a formal ack here?

Gentle reminder.

I'm travelling ATM. Please ping again next week.

Thanks,
Richard

Hi Richard,

Just a kind reminder.

Thanks!

Carolina

maintainer
#6 lore raw

Re: [PATCH net-next V2 1/3] ptp: Add ioctl commands to expose raw cycle counter values

Quoted 6 lines · Carolina Jubran
Quoted 4 lines · Richard Cochran
Quoted 3 lines · Mark Bloch
Quoted 1 line · Paolo Abeni

can we have a formal ack here?

Looks good to me.

Thanks,
Richard

#7 lore raw

Re: [PATCH net-next V2 1/3] ptp: Add ioctl commands to expose raw cycle counter values

Quoted 5 lines · Richard Cochran
Quoted 4 lines · Carolina Jubran
Quoted 2 lines · Richard Cochran
Quoted 1 line · Mark Bloch
Quoted 8 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines · Paolo Abeni
Quoted 1 line

can we have a formal ack here?

Looks good to me.

Thanks,
Richard

Hi Paolo,
A kind reminder, we got the needed ack here.
Yet patchwork state is still 'Needs ACK'.

Regards,
Tariq

maintainer
#8 lore raw

Re: [PATCH net-next V2 1/3] ptp: Add ioctl commands to expose raw cycle counter values

Quoted 11 lines · Tariq Toukan
Quoted 10 lines · Richard Cochran
Quoted 6 lines · Carolina Jubran
Quoted 4 lines · Richard Cochran
Quoted 3 lines · Mark Bloch
Quoted 1 line · Paolo Abeni

can we have a formal ack here?

Looks good to me.

Quoted 7 lines
Quoted 3 lines

Thanks,
Richard

Hi Paolo,
A kind reminder, we got the needed ack here.
Yet patchwork state is still 'Needs ACK'.

I was traveling in the past days.

I see Jakub has some perplexity WRT this series. I read that as "I prefer don't touch it" (as opposed to a "no-go") and I think that the feature makes sense.

I'm applying this now.

/P

maintainer Acked-by on v2 1/3
#9 lore raw

Re: [PATCH net-next V2 1/3] ptp: Add ioctl commands to expose raw cycle counter values

In reply to Tariq Toukan
Quoted 6 lines · Tariq Toukan

From: Carolina Jubran <[email protected]>

Introduce two new ioctl commands, PTP_SYS_OFFSET_PRECISE_CYCLES and PTP_SYS_OFFSET_EXTENDED_CYCLES, to allow user space to access the raw free-running cycle counter from PTP devices.

Quoted 7 lines

These ioctls are variants of the existing PRECISE and EXTENDED offset queries, but instead of returning device time in realtime, they return the raw cycle counter value.

author maintainer
#10 lore raw

PATCH v2 2/3 net/mlx5: Extract MTCTR register read logic into helper function

In reply to Tariq Toukan

From: Carolina Jubran <[email protected]>

Refactor the MTCTR register reading logic into a dedicated helper to lay the groundwork for the next patch.

---
 .../ethernet/mellanox/mlx5/core/lib/clock.c   | 39 +++++++++++++------
 1 file changed, 27 insertions(+), 12 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/lib/clock.c +27 -12
index 214d732d18e9..9b49bdc339ad 100644
@@ -247,27 +247,24 @@ static bool mlx5_is_ptm_source_time_available(struct mlx5_core_dev *dev)
 	return !!MLX5_GET(mtptm_reg, out, psta);
 }
 
-static int mlx5_mtctr_syncdevicetime(ktime_t *device_time,
-				     struct system_counterval_t *sys_counterval,
-				     void *ctx)
+static int mlx5_mtctr_read(struct mlx5_core_dev *mdev,
+			   bool real_time_mode,
+			   struct system_counterval_t *sys_counterval,
+			   u64 *device)
 {
 	u32 out[MLX5_ST_SZ_DW(mtctr_reg)] = {0};
 	u32 in[MLX5_ST_SZ_DW(mtctr_reg)] = {0};
-	struct mlx5_core_dev *mdev = ctx;
-	bool real_time_mode;
-	u64 host, device;
+	u64 host;
 	int err;
 
-	real_time_mode = mlx5_real_time_mode(mdev);
-
 	MLX5_SET(mtctr_reg, in, first_clock_timestamp_request,
 		 MLX5_MTCTR_REQUEST_PTM_ROOT_CLOCK);
 	MLX5_SET(mtctr_reg, in, second_clock_timestamp_request,
 		 real_time_mode ? MLX5_MTCTR_REQUEST_REAL_TIME_CLOCK :
-		 MLX5_MTCTR_REQUEST_FREE_RUNNING_COUNTER);
+				  MLX5_MTCTR_REQUEST_FREE_RUNNING_COUNTER);
 
-	err = mlx5_core_access_reg(mdev, in, sizeof(in), out, sizeof(out), MLX5_REG_MTCTR,
-				   0, 0);
+	err = mlx5_core_access_reg(mdev, in, sizeof(in), out, sizeof(out),
+				   MLX5_REG_MTCTR, 0, 0);
 	if (err)
 		return err;
 
@@ -281,8 +278,26 @@ static int mlx5_mtctr_syncdevicetime(ktime_t *device_time,
 			.cs_id = CSID_X86_ART,
 			.use_nsecs = true,
 	};
+	*device = MLX5_GET64(mtctr_reg, out, second_clock_timestamp);
+
+	return 0;
+}
+
+static int mlx5_mtctr_syncdevicetime(ktime_t *device_time,
+				     struct system_counterval_t *sys_counterval,
+				     void *ctx)
+{
+	struct mlx5_core_dev *mdev = ctx;
+	bool real_time_mode;
+	u64 device;
+	int err;
+
+	real_time_mode = mlx5_real_time_mode(mdev);
+
+	err = mlx5_mtctr_read(mdev, real_time_mode, sys_counterval, &device);
+	if (err)
+		return err;
 
-	device = MLX5_GET64(mtctr_reg, out, second_clock_timestamp);
 	if (real_time_mode)
 		*device_time = ns_to_ktime(REAL_TIME_TO_NS(device >> 32, device & U32_MAX));
 	else
Signature · 1 line
2.40.1
author maintainer
#11 lore raw

PATCH v2 3/3 net/mlx5: Support getcyclesx and getcrosscycles

In reply to Tariq Toukan

From: Carolina Jubran <[email protected]>

Implement the getcyclesx64 and getcrosscycles callbacks in ptp_info to expose the device’s raw free-running counter.

---
 .../ethernet/mellanox/mlx5/core/lib/clock.c   | 74 ++++++++++++++++++-
 1 file changed, 73 insertions(+), 1 deletion(-)
drivers/net/ethernet/mellanox/mlx5/core/lib/clock.c +73 -1
index 9b49bdc339ad..7ad3baca99de 100644
@@ -306,6 +306,23 @@ static int mlx5_mtctr_syncdevicetime(ktime_t *device_time,
 	return 0;
 }
 
+static int
+mlx5_mtctr_syncdevicecyclestime(ktime_t *device_time,
+				struct system_counterval_t *sys_counterval,
+				void *ctx)
+{
+	struct mlx5_core_dev *mdev = ctx;
+	u64 device;
+	int err;
+
+	err = mlx5_mtctr_read(mdev, false, sys_counterval, &device);
+	if (err)
+		return err;
+	*device_time = ns_to_ktime(device);
+
+	return 0;
+}
+
 static int mlx5_ptp_getcrosststamp(struct ptp_clock_info *ptp,
 				   struct system_device_crosststamp *cts)
 {
@@ -330,6 +347,32 @@ static int mlx5_ptp_getcrosststamp(struct ptp_clock_info *ptp,
 	mlx5_clock_unlock(clock);
 	return err;
 }
+
+static int mlx5_ptp_getcrosscycles(struct ptp_clock_info *ptp,
+				   struct system_device_crosststamp *cts)
+{
+	struct mlx5_clock *clock =
+		container_of(ptp, struct mlx5_clock, ptp_info);
+	struct system_time_snapshot history_begin = {0};
+	struct mlx5_core_dev *mdev;
+	int err;
+
+	mlx5_clock_lock(clock);
+	mdev = mlx5_clock_mdev_get(clock);
+
+	if (!mlx5_is_ptm_source_time_available(mdev)) {
+		err = -EBUSY;
+		goto unlock;
+	}
+
+	ktime_get_snapshot(&history_begin);
+
+	err = get_device_system_crosststamp(mlx5_mtctr_syncdevicecyclestime,
+					    mdev, &history_begin, cts);
+unlock:
+	mlx5_clock_unlock(clock);
+	return err;
+}
 #endif /* CONFIG_X86 */
 
 static u64 mlx5_read_time(struct mlx5_core_dev *dev,
@@ -528,6 +571,24 @@ static int mlx5_ptp_gettimex(struct ptp_clock_info *ptp, struct timespec64 *ts,
 	return 0;
 }
 
+static int mlx5_ptp_getcyclesx(struct ptp_clock_info *ptp,
+			       struct timespec64 *ts,
+			       struct ptp_system_timestamp *sts)
+{
+	struct mlx5_clock *clock = container_of(ptp, struct mlx5_clock,
+						ptp_info);
+	struct mlx5_core_dev *mdev;
+	u64 cycles;
+
+	mlx5_clock_lock(clock);
+	mdev = mlx5_clock_mdev_get(clock);
+
+	cycles = mlx5_read_time(mdev, sts, false);
+	*ts = ns_to_timespec64(cycles);
+	mlx5_clock_unlock(clock);
+	return 0;
+}
+
 static int mlx5_ptp_adjtime_real_time(struct mlx5_core_dev *mdev, s64 delta)
 {
 	u32 in[MLX5_ST_SZ_DW(mtutc_reg)] = {};
@@ -1244,6 +1305,7 @@ static void mlx5_init_timer_max_freq_adjustment(struct mlx5_core_dev *mdev)
 static void mlx5_init_timer_clock(struct mlx5_core_dev *mdev)
 {
 	struct mlx5_clock *clock = mdev->clock;
+	bool expose_cycles;
 
 	/* Configure the PHC */
 	clock->ptp_info = mlx5_ptp_clock_info;
@@ -1251,12 +1313,22 @@ static void mlx5_init_timer_clock(struct mlx5_core_dev *mdev)
 	if (MLX5_CAP_MCAM_REG(mdev, mtutc))
 		mlx5_init_timer_max_freq_adjustment(mdev);
 
+	expose_cycles = !MLX5_CAP_GEN(mdev, disciplined_fr_counter) ||
+			!mlx5_real_time_mode(mdev);
+
 #ifdef CONFIG_X86
 	if (MLX5_CAP_MCAM_REG3(mdev, mtptm) &&
-	    MLX5_CAP_MCAM_REG3(mdev, mtctr) && boot_cpu_has(X86_FEATURE_ART))
+	    MLX5_CAP_MCAM_REG3(mdev, mtctr) && boot_cpu_has(X86_FEATURE_ART)) {
 		clock->ptp_info.getcrosststamp = mlx5_ptp_getcrosststamp;
+		if (expose_cycles)
+			clock->ptp_info.getcrosscycles =
+				mlx5_ptp_getcrosscycles;
+	}
 #endif /* CONFIG_X86 */
 
+	if (expose_cycles)
+		clock->ptp_info.getcyclesx64 = mlx5_ptp_getcyclesx;
+
 	mlx5_timecounter_init(mdev);
 	mlx5_init_clock_info(mdev);
 	mlx5_init_overflow_period(mdev);
Signature · 1 line
2.40.1
#12 lore raw
In reply to Tariq Toukan
This series was applied to netdev/net-next.git (main) by Paolo Abeni <[email protected]>:

Hello:

This series was applied to netdev/net-next.git (main) by Paolo Abeni <[email protected]>:

Quoted 9 lines · On Tue, 12 Aug 2025 17:17:05 +0300 you wrote:

Hi,

This series by Carolina adds support in ptp and usage in mlx5 for exposing the raw free-running cycle counter of PTP hardware clocks.

This is V2. Find previous one here:
https://lore.kernel.org/all/[email protected]/

[...]

Here is the summary with links:
- [net-next,V2,1/3] ptp: Add ioctl commands to expose raw cycle counter values

    https://git.kernel.org/netdev/net-next/c/faf23f54d366
  - [net-next,V2,2/3] net/mlx5: Extract MTCTR register read logic into helper function
    https://git.kernel.org/netdev/net-next/c/96c345c3c54c
  - [net-next,V2,3/3] net/mlx5: Support getcyclesx and getcrosscycles
    https://git.kernel.org/netdev/net-next/c/a3fb485505ca

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html