Re: [PATCH v3 1/9] drm/xe/hwmon: initialize fan-control backend and table cache

["Purkait, Soham" <soham.purkait@intel.com>]

[-- Attachment #1: Type: text/plain, Size: 10971 bytes --]

Hi Karthik,

On 16-10-2026 00:37, Karthik Poosa wrote:
> Initialize Xe hwmon fan-control support by detecting fan count,
> reading stock fan control points and min PWM.
>
> v2:
>   - Avoid user table initialization during probe. (Badal)
>   - Move unused code to appropriate patches.
>   - Use xe helpers for dmesg logs.
>
> v3:
>   - Rebase.
>   - Refine a few debug log messages and comments.
>   - Update xe_hwmon_read_fan_control_info() to return on any stock table
>     read failures.
>   - Avoid returning when num_fans is 0 in
>     xe_hwmon_get_preregistration_info(), ensuring thermal mailbox
>     initialization is not skipped.
>   - Remove XE_PCODE_FAN_CONTROL_TIMEOUT_MS which is unncessary as default
>     timeout itself is now 10ms.
>   - Use xe_pcode_read() instead of xe_pcode_read_timeout().
>   - Rename fan_control_point_count to point_count in struct fan_table.
>   - Resolve kgdoc warnings.
>   - Add valid check for stock table fan point count.
>   - Use the fan-enable bitmap reported by FSC_READ_NUM_FANS when enumerating
>     fans. This avoids assuming contiguous fan indices and allows masks with
>     gaps to be represented correctly in the hwmon interface.
>
> Signed-off-by: Karthik Poosa<karthik.poosa@intel.com>
> Assisted-by: LLM
> ---
>   drivers/gpu/drm/xe/xe_hwmon.c     | 167 +++++++++++++++++++++++++++---
>   drivers/gpu/drm/xe/xe_pcode_api.h |   5 +
>   2 files changed, 157 insertions(+), 15 deletions(-)
>
> diff --git a/drivers/gpu/drm/xe/xe_hwmon.c b/drivers/gpu/drm/xe/xe_hwmon.c
> index 42acfeb8818c..f2bd20f8b0a2 100644
> --- a/drivers/gpu/drm/xe/xe_hwmon.c
> +++ b/drivers/gpu/drm/xe/xe_hwmon.c
> @@ -21,6 +21,7 @@
>   #include "xe_pcode_api.h"
>   #include "xe_sriov.h"
>   #include "xe_pm.h"
> +#include "xe_printk.h"
>   #include "xe_vsec.h"
>   #include "regs/xe_pmt.h"
>   
> @@ -137,14 +138,51 @@ struct xe_hwmon_energy_info {
>   	long accum_energy;
>   };
>   
> +enum fan_table_type {
> +	/** @STOCK_FAN_TABLE: firmware-provided fan table */
> +	STOCK_FAN_TABLE,
> +	/** @USER_FAN_TABLE: user fan table written through sysfs */
> +	USER_FAN_TABLE,
> +	/** @FAN_TABLE_MAX: number of fan tables tracked per fan */
> +	FAN_TABLE_MAX,
> +};
> +
> +/* Check if a fan is enabled based on the fan_enable_mask got from FSC_READ_NUM_FANS. */
> +#define IS_FAN_ENABLED(fan) (hwmon->fan_enable_mask & BIT(fan))
If this macro is called from any function having no local variable named 
hwmon it will break. Should it take hwmon as an argument ?
> +
> +/* Maximum number of fan control points supported by each fan. */
> +#define MAX_FAN_CONTROL_POINTS	(10)
> +
> +/* Fan control point index: bit31 selects fan table type; bits30:0 control point index. */
> +#define FCP_INDEX(FAN_TABLE_TYPE, POINT_NUM) (((FAN_TABLE_TYPE) << 31) | \
Cast to u32 ? shifting the signed enum into bit 31 could lead to UB.
> +						((POINT_NUM) & REG_GENMASK(30, 0)))
> +
>   /**
> - * struct xe_hwmon_fan_info - to cache previous fan reading
> + * struct xe_hwmon_fan_info - cached fan telemetry and control state
> + *
> + * Each fan keeps the latest tachometer sampling state along with two fan
> + * tables: the stock table discovered from firmware and the user table managed
> + * by hwmon sysfs writes.
>    */
>   struct xe_hwmon_fan_info {
>   	/** @reg_val_prev: previous fan reg val */
>   	u32 reg_val_prev;
>   	/** @time_prev: previous timestamp */
>   	u64 time_prev;
> +	/** @fan_table: fan control tables */
> +	struct fan_table {
> +		/** @point_count: number of fan control points */
> +		u8 point_count;
> +		/** @fcp: array of fan control points */
> +		struct fan_control_point {
> +			/** @temp: temperature in degree celsius */
> +			u8 temp;
> +			/** @speed: fan speed in percentage */
> +			u8 speed;
> +		} fcp[MAX_FAN_CONTROL_POINTS];
> +	} fan_table[FAN_TABLE_MAX];
> +	/** @min_pwm: minimum fan PWM */
> +	u32 min_pwm;
>   };
>   
>   /**
> @@ -193,6 +231,8 @@ struct xe_hwmon {
>   	int scl_shift_time;
>   	/** @ei: Energy info for energyN_input */
>   	struct xe_hwmon_energy_info ei[CHANNEL_MAX];
> +	/** @fan_enable_mask: bitmask of available fans */
> +	u32 fan_enable_mask;
>   	/** @fi: Fan info for fanN_input */
>   	struct xe_hwmon_fan_info fi[FAN_MAX];
>   	/** @read_power_limit_from_pcode: pcode accepted power limits write for the channel */
> @@ -1075,17 +1115,112 @@ static int xe_hwmon_pcode_write_i1(const struct xe_hwmon *hwmon, u32 uval)
>   			      (uval & POWER_SETUP_I1_DATA_MASK));
>   }
>   
> -static int xe_hwmon_pcode_read_fan_control(const struct xe_hwmon *hwmon, u32 subcmd, u32 *uval)
> +static int xe_hwmon_pcode_read_fan_control(const struct xe_hwmon *hwmon, u32 subcmd, u8 fan_num,
> +					   u32 *uval)
>   {
>   	struct xe_tile *root_tile = xe_device_get_root_tile(hwmon->xe);
Nit: blank line after variable declaration
> +	return xe_pcode_read(root_tile, PCODE_MBOX(FAN_SPEED_CONTROL, subcmd, fan_num),
> +				     uval, NULL);
> +}
> +
> +static int xe_hwmon_get_fan_enable_mask(struct xe_hwmon *hwmon)
> +{
> +	int ret;
>   
>   	/* Platforms that don't return correct value */
> -	if (hwmon->xe->info.platform == XE_DG2 && subcmd == FSC_READ_NUM_FANS) {
> -		*uval = 2;
> +	if (hwmon->xe->info.platform == XE_DG2) {
> +		hwmon->fan_enable_mask = GENMASK(1, 0);
>   		return 0;
>   	}
>   
> -	return xe_pcode_read(root_tile, PCODE_MBOX(FAN_SPEED_CONTROL, subcmd, 0), uval, NULL);
> +	ret = xe_hwmon_pcode_read_fan_control(hwmon, FSC_READ_NUM_FANS, 0,
> +					      &hwmon->fan_enable_mask);
> +	if (ret) {
> +		xe_warn(hwmon->xe, "failed to read number of fans, ret=%d\n", ret);
> +		return ret;
> +	}
> +
> +	/* Check if fan mask exceeds FAN_MAX and log a warning if it does */
> +	if (hwmon->fan_enable_mask & ~GENMASK(FAN_MAX - 1, 0)) {
> +		xe_warn(hwmon->xe, "fan enable mask exceeds maximum supported fans\n");
Reset the mask to 0 so IS_FAN_ENABLED() stays false in case fan mask 
exceeds FAN_MAX.
> +		return -EINVAL;
> +	}
> +
> +	xe_info(hwmon->xe, "Number of fans detected: %u\n",
> +		hweight32(hwmon->fan_enable_mask));
> +	return 0;
> +}
> +
> +static int xe_hwmon_read_fan_control_info(struct xe_hwmon *hwmon)
> +{
> +	struct xe_tile *root_tile = xe_device_get_root_tile(hwmon->xe);
> +	int point;
> +	int fan;
> +	int ret;
> +
> +	for (fan = 0; fan < FAN_MAX; fan++) {
> +		u32 stock_fcp_count = 0;
> +
> +		struct xe_hwmon_fan_info *fi = &hwmon->fi[fan];
> +
> +		if (!IS_FAN_ENABLED(fan))
> +			continue;
> +
> +		ret = xe_hwmon_pcode_read_fan_control(hwmon,
> +						      FSC_READ_STOCK_FAN_CONTROL_POINTS,
> +						      fan, &stock_fcp_count);
> +		if (ret) {
> +			xe_err(hwmon->xe,
> +			       "failed to read fan %d stock table point count, ret=%d\n",
> +			       fan, ret);
Why not continue for the rest of the fans marking the current one ?
> +			return ret;
> +		}
> +
> +		if (!stock_fcp_count || stock_fcp_count > MAX_FAN_CONTROL_POINTS) {
> +			xe_warn(hwmon->xe,
> +				"fan %d stock table point count %u is invalid\n",
> +				fan, stock_fcp_count);
Ditto.
> +			return -EINVAL;
> +		}
> +
> +		fi->fan_table[STOCK_FAN_TABLE].point_count = stock_fcp_count;
> +		xe_dbg(hwmon->xe, "fan %d stock table point count is %u\n", fan,
> +		       fi->fan_table[STOCK_FAN_TABLE].point_count);
> +
> +		/* Cache the stock fan table for later use. */
> +		for (point = 0; point < fi->fan_table[STOCK_FAN_TABLE].point_count;
> +		     point++) {
> +			u32 fcp = 0;
> +
> +			fcp = FCP_INDEX(STOCK_FAN_TABLE, point);

nit: u32 fcp = FCP_INDEX(STOCK_FAN_TABLE, point);

Regards,
Soham
> +			ret = xe_pcode_read(root_tile, PCODE_MBOX(FAN_SPEED_CONTROL,
> +								  FSC_READ_FAN_TABLE, fan),
> +					    &fcp, NULL);
> +			if (ret) {
> +				xe_err(hwmon->xe, "failed to read fan %d stock point %d, ret=%d\n",
> +				       fan, point, ret);
> +				fi->fan_table[STOCK_FAN_TABLE].point_count = 0;
> +				return ret;
> +			}
> +
> +			fi->fan_table[STOCK_FAN_TABLE].fcp[point].temp =
> +						REG_FIELD_GET(FAN_CONTROL_POINT_TEMP_MASK, fcp);
> +			fi->fan_table[STOCK_FAN_TABLE].fcp[point].speed =
> +						REG_FIELD_GET(FAN_CONTROL_POINT_SPEED_MASK, fcp);
> +			xe_dbg(hwmon->xe, "fan %d stock point %d: temp %u C, speed %u\n",
> +			       fan, point, fi->fan_table[STOCK_FAN_TABLE].fcp[point].temp,
> +			       fi->fan_table[STOCK_FAN_TABLE].fcp[point].speed);
> +		}
> +
> +		/* Read minimum fan PWM */
> +		ret = xe_hwmon_pcode_read_fan_control(hwmon, FSC_READ_FAN_MIN_PWM, fan,
> +						      &fi->min_pwm);
> +		if (ret)
> +			xe_warn(hwmon->xe, "fan %d min pwm unavailable, ret=%d\n", fan, ret);
> +		else
> +			xe_dbg(hwmon->xe, "fan %d min PWM %u\n", fan, fi->min_pwm);
> +	}
> +	return 0;
>   }
>   
>   static int xe_hwmon_power_curr_crit_read(struct xe_hwmon *hwmon, int channel,
> @@ -1615,17 +1750,12 @@ xe_hwmon_energy_read(struct xe_hwmon *hwmon, u32 attr, int channel, long *val)
>   static umode_t
>   xe_hwmon_fan_is_visible(struct xe_hwmon *hwmon, u32 attr, int channel)
>   {
> -	u32 uval = 0;
> -
>   	if (!hwmon->xe->info.has_fan_control)
>   		return 0;
>   
>   	switch (attr) {
>   	case hwmon_fan_input:
> -		if (xe_hwmon_pcode_read_fan_control(hwmon, FSC_READ_NUM_FANS, &uval))
> -			return 0;
> -
> -		return channel < uval ? 0444 : 0;
> +		return IS_FAN_ENABLED(channel) ? 0444 : 0;
>   	default:
>   		return 0;
>   	}
> @@ -1858,11 +1988,18 @@ xe_hwmon_get_preregistration_info(struct xe_hwmon *hwmon)
>   		if (xe_hwmon_is_visible(hwmon, hwmon_energy, hwmon_energy_input, channel))
>   			xe_hwmon_energy_get(hwmon, channel, &energy);
>   
> -	/* Initialize 'struct xe_hwmon_fan_info' with initial fan register reading. */
> -	for (channel = 0; channel < FAN_MAX; channel++)
> -		if (xe_hwmon_is_visible(hwmon, hwmon_fan, hwmon_fan_input, channel))
> -			xe_hwmon_fan_input_read(hwmon, channel, &fan_speed);
> +	if (hwmon->xe->info.has_fan_control) {
> +		xe_hwmon_get_fan_enable_mask(hwmon);
> +
> +		/* Initialize 'struct xe_hwmon_fan_info' with initial fan register reading. */
> +		for (channel = 0; channel < FAN_MAX; channel++)
> +			if (IS_FAN_ENABLED(channel))
> +				xe_hwmon_fan_input_read(hwmon, channel, &fan_speed);
>   
> +		/* Fan control tables initialization */
> +		if (xe_hwmon_read_fan_control_info(hwmon))
> +			xe_warn(hwmon->xe, "Fan control tables are not available\n");
> +	}
>   	return xe_hwmon_init_temp_info(hwmon);
>   }
>   
> diff --git a/drivers/gpu/drm/xe/xe_pcode_api.h b/drivers/gpu/drm/xe/xe_pcode_api.h
> index 00c4fc7d0124..0349b24a8259 100644
> --- a/drivers/gpu/drm/xe/xe_pcode_api.h
> +++ b/drivers/gpu/drm/xe/xe_pcode_api.h
> @@ -83,6 +83,11 @@
>   
>   #define   FAN_SPEED_CONTROL			0x7D
>   #define     FSC_READ_NUM_FANS			0x4
> +#define     FSC_READ_STOCK_FAN_CONTROL_POINTS	0x5
> +#define     FSC_READ_FAN_TABLE			0x7
> +#define     FAN_CONTROL_POINT_TEMP_MASK		REG_GENMASK(7, 0)
> +#define     FAN_CONTROL_POINT_SPEED_MASK	REG_GENMASK(15, 8)
> +#define     FSC_READ_FAN_MIN_PWM		0x8
>   
>   #define PCODE_SCRATCH(x)		XE_REG(0x138320 + ((x) * 4))
>   /* PCODE_SCRATCH0 */

[-- Attachment #2: Type: text/html, Size: 12454 bytes --]