Merge branch 'mlx5-ptm-cross-timestamping-support'

Tariq Toukan says:

====================
mlx5 PTM cross timestamping support

This patchset by Rahul and Carolina adds PTM (Precision Time Measurement)
support to the mlx5 driver.

PTM is a PCI extended capability introduced by PCI-SIG for providing an
accurate read of the device clock offset without being impacted by
asymmetric bus transfer rates.

The performance of PTM on ConnectX-7 was evaluated using both real-time
(RTC) and free-running (FRC) clocks under traffic and no traffic
conditions. Tests with phc2sys measured the maximum offset values at a 50Hz
rate, with and without PTM.

Results:

1. No traffic
+-----+--------+--------+
|     | No-PTM | PTM    |
+-----+--------+--------+
| FRC | 125 ns | <29 ns |
+-----+--------+--------+
| RTC | 248 ns | <34 ns |
+-----+--------+--------+

2. With traffic
+-----+--------+--------+
|     | No-PTM | PTM    |
+-----+--------+--------+
| FRC | 254 ns | <40 ns |
+-----+--------+--------+
| RTC | 255 ns | <45 ns |
+-----+--------+--------+

v2: https://lore.kernel.org/[email protected]
====================

Link: https://patch.msgid.link/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

Author: kuba-moo

drivers/net/ethernet/mellanox/mlx5/core/fw.c

@@ -224,6 +224,7 @@ int mlx5_query_hca_caps(struct mlx5_core_dev *dev)
 	if (MLX5_CAP_GEN(dev, mcam_reg)) {
 		mlx5_get_mcam_access_reg_group(dev, MLX5_MCAM_REGS_FIRST_128);
 		mlx5_get_mcam_access_reg_group(dev, MLX5_MCAM_REGS_0x9100_0x917F);
+		mlx5_get_mcam_access_reg_group(dev, MLX5_MCAM_REGS_0x9180_0x91FF);
 	}
 
 	if (MLX5_CAP_GEN(dev, qcam_reg))

drivers/net/ethernet/mellanox/mlx5/core/lib/clock.c

@@ -38,6 +38,10 @@
 #include "lib/eq.h"
 #include "en.h"
 #include "clock.h"
+#ifdef CONFIG_X86
+#include <linux/timekeeping.h>
+#include <linux/cpufeature.h>
+#endif /* CONFIG_X86 */
 
 enum {
 	MLX5_PIN_MODE_IN		= 0x0,
@@ -148,6 +152,87 @@ static int mlx5_set_mtutc(struct mlx5_core_dev *dev, u32 *mtutc, u32 size)
 				    MLX5_REG_MTUTC, 0, 1);
 }
 
+#ifdef CONFIG_X86
+static bool mlx5_is_ptm_source_time_available(struct mlx5_core_dev *dev)
+{
+	u32 out[MLX5_ST_SZ_DW(mtptm_reg)] = {0};
+	u32 in[MLX5_ST_SZ_DW(mtptm_reg)] = {0};
+	int err;
+
+	if (!MLX5_CAP_MCAM_REG3(dev, mtptm))
+		return false;
+
+	err = mlx5_core_access_reg(dev, in, sizeof(in), out, sizeof(out), MLX5_REG_MTPTM,
+				   0, 0);
+	if (err)
+		return false;
+
+	return !!MLX5_GET(mtptm_reg, out, psta);
+}
+
+static int mlx5_mtctr_syncdevicetime(ktime_t *device_time,
+				     struct system_counterval_t *sys_counterval,
+				     void *ctx)
+{
+	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;
+	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);
+
+	err = mlx5_core_access_reg(mdev, in, sizeof(in), out, sizeof(out), MLX5_REG_MTCTR,
+				   0, 0);
+	if (err)
+		return err;
+
+	if (!MLX5_GET(mtctr_reg, out, first_clock_valid) ||
+	    !MLX5_GET(mtctr_reg, out, second_clock_valid))
+		return -EINVAL;
+
+	host = MLX5_GET64(mtctr_reg, out, first_clock_timestamp);
+	*sys_counterval = (struct system_counterval_t) {
+			.cycles = host,
+			.cs_id = CSID_X86_ART,
+			.use_nsecs = true,
+	};
+
+	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
+		*device_time = mlx5_timecounter_cyc2time(&mdev->clock, device);
+
+	return 0;
+}
+
+static int mlx5_ptp_getcrosststamp(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;
+
+	mdev = container_of(clock, struct mlx5_core_dev, clock);
+
+	if (!mlx5_is_ptm_source_time_available(mdev))
+		return -EBUSY;
+
+	ktime_get_snapshot(&history_begin);
+
+	return get_device_system_crosststamp(mlx5_mtctr_syncdevicetime, mdev,
+					     &history_begin, cts);
+}
+#endif /* CONFIG_X86 */
+
 static u64 mlx5_read_time(struct mlx5_core_dev *dev,
 			  struct ptp_system_timestamp *sts,
 			  bool real_time)
@@ -1034,6 +1119,12 @@ 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);
 
+#ifdef CONFIG_X86
+	if (MLX5_CAP_MCAM_REG3(mdev, mtptm) &&
+	    MLX5_CAP_MCAM_REG3(mdev, mtctr) && boot_cpu_has(X86_FEATURE_ART))
+		clock->ptp_info.getcrosststamp = mlx5_ptp_getcrosststamp;
+#endif /* CONFIG_X86 */
+
 	mlx5_timecounter_init(mdev);
 	mlx5_init_clock_info(mdev);
 	mlx5_init_overflow_period(clock);

drivers/net/ethernet/mellanox/mlx5/core/main.c

@@ -923,6 +923,11 @@ static int mlx5_pci_init(struct mlx5_core_dev *dev, struct pci_dev *pdev,
 	}
 
 	mlx5_pci_vsc_init(dev);
+
+	err = pci_enable_ptm(pdev, NULL);
+	if (err)
+		mlx5_core_info(dev, "PTM is not supported by PCIe\n");
+
 	return 0;
 
 err_clr_master:
@@ -939,6 +944,7 @@ static void mlx5_pci_close(struct mlx5_core_dev *dev)
 	 * before removing the pci bars
 	 */
 	mlx5_drain_health_wq(dev);
+	pci_disable_ptm(dev->pdev);
 	iounmap(dev->iseg);
 	release_bar(dev->pdev);
 	mlx5_pci_disable_device(dev);

include/linux/mlx5/device.h

@@ -1243,7 +1243,8 @@ enum mlx5_pcam_feature_groups {
 enum mlx5_mcam_reg_groups {
 	MLX5_MCAM_REGS_FIRST_128                    = 0x0,
 	MLX5_MCAM_REGS_0x9100_0x917F                = 0x2,
-	MLX5_MCAM_REGS_NUM                          = 0x3,
+	MLX5_MCAM_REGS_0x9180_0x91FF                = 0x3,
+	MLX5_MCAM_REGS_NUM                          = 0x4,
 };
 
 enum mlx5_mcam_feature_groups {
@@ -1392,6 +1393,10 @@ enum mlx5_qcam_feature_groups {
 	MLX5_GET(mcam_reg, (mdev)->caps.mcam[MLX5_MCAM_REGS_0x9100_0x917F], \
 		 mng_access_reg_cap_mask.access_regs2.reg)
 
+#define MLX5_CAP_MCAM_REG3(mdev, reg) \
+	MLX5_GET(mcam_reg, (mdev)->caps.mcam[MLX5_MCAM_REGS_0x9180_0x91FF], \
+		 mng_access_reg_cap_mask.access_regs3.reg)
+
 #define MLX5_CAP_MCAM_FEATURE(mdev, fld) \
 	MLX5_GET(mcam_reg, (mdev)->caps.mcam, mng_feature_cap_mask.enhanced_features.fld)
 

include/linux/mlx5/driver.h

@@ -159,6 +159,8 @@ enum {
 	MLX5_REG_MSECQ		 = 0x9155,
 	MLX5_REG_MSEES		 = 0x9156,
 	MLX5_REG_MIRC		 = 0x9162,
+	MLX5_REG_MTPTM		 = 0x9180,
+	MLX5_REG_MTCTR		 = 0x9181,
 	MLX5_REG_SBCAM		 = 0xB01F,
 	MLX5_REG_RESOURCE_DUMP   = 0xC000,
 	MLX5_REG_DTOR            = 0xC00E,

include/linux/mlx5/mlx5_ifc.h

@@ -10401,6 +10401,18 @@ struct mlx5_ifc_mcam_access_reg_bits2 {
 	u8         regs_31_to_0[0x20];
 };
 
+struct mlx5_ifc_mcam_access_reg_bits3 {
+	u8         regs_127_to_96[0x20];
+
+	u8         regs_95_to_64[0x20];
+
+	u8         regs_63_to_32[0x20];
+
+	u8         regs_31_to_2[0x1e];
+	u8         mtctr[0x1];
+	u8         mtptm[0x1];
+};
+
 struct mlx5_ifc_mcam_reg_bits {
 	u8         reserved_at_0[0x8];
 	u8         feature_group[0x8];
@@ -10413,6 +10425,7 @@ struct mlx5_ifc_mcam_reg_bits {
 		struct mlx5_ifc_mcam_access_reg_bits access_regs;
 		struct mlx5_ifc_mcam_access_reg_bits1 access_regs1;
 		struct mlx5_ifc_mcam_access_reg_bits2 access_regs2;
+		struct mlx5_ifc_mcam_access_reg_bits3 access_regs3;
 		u8         reserved_at_0[0x80];
 	} mng_access_reg_cap_mask;
 
@@ -11166,6 +11179,34 @@ struct mlx5_ifc_mtmp_reg_bits {
 	u8         sensor_name_lo[0x20];
 };
 
+struct mlx5_ifc_mtptm_reg_bits {
+	u8         reserved_at_0[0x10];
+	u8         psta[0x1];
+	u8         reserved_at_11[0xf];
+
+	u8         reserved_at_20[0x60];
+};
+
+enum {
+	MLX5_MTCTR_REQUEST_NOP = 0x0,
+	MLX5_MTCTR_REQUEST_PTM_ROOT_CLOCK = 0x1,
+	MLX5_MTCTR_REQUEST_FREE_RUNNING_COUNTER = 0x2,
+	MLX5_MTCTR_REQUEST_REAL_TIME_CLOCK = 0x3,
+};
+
+struct mlx5_ifc_mtctr_reg_bits {
+	u8         first_clock_timestamp_request[0x8];
+	u8         second_clock_timestamp_request[0x8];
+	u8         reserved_at_10[0x10];
+
+	u8         first_clock_valid[0x1];
+	u8         second_clock_valid[0x1];
+	u8         reserved_at_22[0x1e];
+
+	u8         first_clock_timestamp[0x40];
+	u8         second_clock_timestamp[0x40];
+};
+
 union mlx5_ifc_ports_control_registers_document_bits {
 	struct mlx5_ifc_bufferx_reg_bits bufferx_reg;
 	struct mlx5_ifc_eth_2819_cntrs_grp_data_layout_bits eth_2819_cntrs_grp_data_layout;
@@ -11230,6 +11271,8 @@ union mlx5_ifc_ports_control_registers_document_bits {
 	struct mlx5_ifc_mrtc_reg_bits mrtc_reg;
 	struct mlx5_ifc_mtcap_reg_bits mtcap_reg;
 	struct mlx5_ifc_mtmp_reg_bits mtmp_reg;
+	struct mlx5_ifc_mtptm_reg_bits mtptm_reg;
+	struct mlx5_ifc_mtctr_reg_bits mtctr_reg;
 	u8         reserved_at_0[0x60e0];
 };