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

   1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
   2 // Copyright (c) 2023, NVIDIA CORPORATION & AFFILIATES.
   3
   4 #include <linux/kernel.h>
   5 #include <linux/types.h>
   6 #include <linux/device.h>
   7 #include <linux/thermal.h>
   8 #include <linux/err.h>
   9 #include <linux/mlx5/driver.h>
  10 #include "mlx5_core.h"
  11 #include "thermal.h"
  12
  13 #define MLX5_THERMAL_POLL_INT_MSEC      1000
  14 #define MLX5_THERMAL_NUM_TRIPS          0
  15 #define MLX5_THERMAL_ASIC_SENSOR_INDEX  0
  16
  17 /* Bit string indicating the writeablility of trip points if any */
  18 #define MLX5_THERMAL_TRIP_MASK  (BIT(MLX5_THERMAL_NUM_TRIPS) - 1)
  19
  20 struct mlx5_thermal {
  21         struct mlx5_core_dev *mdev;
  22         struct thermal_zone_device *tzdev;
  23 };
  24
  25 static int mlx5_thermal_get_mtmp_temp(struct mlx5_core_dev *mdev, u32 id, int *p_temp)
  26 {
  27         u32 mtmp_out[MLX5_ST_SZ_DW(mtmp_reg)] = {};
  28         u32 mtmp_in[MLX5_ST_SZ_DW(mtmp_reg)] = {};
  29         int err;
  30
  31         MLX5_SET(mtmp_reg, mtmp_in, sensor_index, id);
  32
  33         err = mlx5_core_access_reg(mdev, mtmp_in,  sizeof(mtmp_in),
  34                                    mtmp_out, sizeof(mtmp_out),
  35                                    MLX5_REG_MTMP, 0, 0);
  36
  37         if (err)
  38                 return err;
  39
  40         *p_temp = MLX5_GET(mtmp_reg, mtmp_out, temperature);
  41
  42         return 0;
  43 }
  44
  45 static int mlx5_thermal_get_temp(struct thermal_zone_device *tzdev,
  46                                  int *p_temp)
  47 {
  48         struct mlx5_thermal *thermal = thermal_zone_device_priv(tzdev);
  49         struct mlx5_core_dev *mdev = thermal->mdev;
  50         int err;
  51
  52         err = mlx5_thermal_get_mtmp_temp(mdev, MLX5_THERMAL_ASIC_SENSOR_INDEX, p_temp);
  53
  54         if (err)
  55                 return err;
  56
  57         /* The unit of temp returned is in 0.125 C. The thermal
  58          * framework expects the value in 0.001 C.
  59          */
  60         *p_temp *= 125;
  61
  62         return 0;
  63 }
  64
  65 static struct thermal_zone_device_ops mlx5_thermal_ops = {
  66         .get_temp = mlx5_thermal_get_temp,
  67 };
  68
  69 int mlx5_thermal_init(struct mlx5_core_dev *mdev)
  70 {
  71         char data[THERMAL_NAME_LENGTH];
  72         struct mlx5_thermal *thermal;
  73         int err;
  74
  75         if (!mlx5_core_is_pf(mdev) && !mlx5_core_is_ecpf(mdev))
  76                 return 0;
  77
  78         err = snprintf(data, sizeof(data), "mlx5_%s", dev_name(mdev->device));
  79         if (err < 0 || err >= sizeof(data)) {
  80                 mlx5_core_err(mdev, "Failed to setup thermal zone name, %d\n", err);
  81                 return -EINVAL;
  82         }
  83
  84         thermal = kzalloc(sizeof(*thermal), GFP_KERNEL);
  85         if (!thermal)
  86                 return -ENOMEM;
  87
  88         thermal->mdev = mdev;
  89         thermal->tzdev = thermal_zone_device_register_with_trips(data,
  90                                                                  NULL,
  91                                                                  MLX5_THERMAL_NUM_TRIPS,
  92                                                                  MLX5_THERMAL_TRIP_MASK,
  93                                                                  thermal,
  94                                                                  &mlx5_thermal_ops,
  95                                                                  NULL, 0, MLX5_THERMAL_POLL_INT_MSEC);
  96         if (IS_ERR(thermal->tzdev)) {
  97                 err = PTR_ERR(thermal->tzdev);
  98                 mlx5_core_err(mdev, "Failed to register thermal zone device (%s) %d\n", data, err);
  99                 kfree(thermal);
 100                 return err;
 101         }
 102
 103         mdev->thermal = thermal;
 104         return 0;
 105 }
 106
 107 void mlx5_thermal_uninit(struct mlx5_core_dev *mdev)
 108 {
 109         if (!mdev->thermal)
 110                 return;
 111
 112         thermal_zone_device_unregister(mdev->thermal->tzdev);
 113         kfree(mdev->thermal);
 114 }