Re: [PATCH 04/11] scsi: ufs: core: Add support for TX Equalization

[Can Guo <can.guo@oss.qualcomm.com>]

Hi Bart,

On 2/28/2026 5:39 AM, Bart Van Assche wrote:
>
> On 2/27/26 8:08 AM, Can Guo wrote:
>> +static bool use_adaptive_txeq;
>> +module_param(use_adaptive_txeq, bool, 0644);
>> +MODULE_PARM_DESC(use_adaptive_txeq, "Find and apply optimal TX 
>> Equalization settings before power mode change (default: false)");
>> +
>> +static int txeq_gear_set(const char *val, const struct kernel_param 
>> *kp)
>> +{
>> +    return param_set_uint_minmax(val, kp, UFS_HS_G1, UFS_HS_G6);
>> +}
>> +
>> +static const struct kernel_param_ops txeq_gear_ops = {
>> +    .set = txeq_gear_set,
>> +    .get = param_get_uint,
>> +};
>> +
>> +static unsigned int adaptive_txeq_gear = UFS_HS_G6;
>> +module_param_cb(adaptive_txeq_gear, &txeq_gear_ops, 
>> &adaptive_txeq_gear, 0644);
>> +MODULE_PARM_DESC(adaptive_txeq_gear, "For the HS-Gear[n] and above, 
>> adaptive txeq shall be used");
>> +
>> +static bool use_txeq_presets = true;
>> +module_param(use_txeq_presets, bool, 0644);
>> +MODULE_PARM_DESC(use_txeq_presets, "Use only the 8 TX Equalization 
>> Presets (pre-defined Pre-Shoot & De-Emphasis combinations) for TX 
>> EQTR (default: true)");
>> +
>> +static bool txeq_presets_selected[UFS_TX_EQ_PRESET_MAX] = {[0 ... 
>> (UFS_TX_EQ_PRESET_MAX - 1)] = 1};
>> +module_param_array(txeq_presets_selected, bool, NULL, 0644);
>> +MODULE_PARM_DESC(txeq_presets_selected, "Use only the selected 
>> Presets out of the 8 TX Equalization Presets for TX EQTR");
>
> Please minimize the kernel module parameters. Introducing new kernel 
> module parameters is easy but removing these is hard. Can all the above
> parameters be removed? If not, please only keep what is absolutely
> necessary.
The three module parameters are necessary -
1. use_adaptive_txeq parameter is needed as an overall disable/enable
    switch.
2. use_txeq_presets parameter is needed because a full TX EQTR procedure
    with all 64 (8 x 8) PreShoot&DeEmphasis combinations give more
    accurate TX EQTR result, but it takes long time. use_txeq_presets
    parameter selects only the 8 combinations (or presets defined by M-PHY
    spec) as a tradeoff.
3. txeq_presets_selected parameter provides a finer resolution of the 8
    presets to further reduce TX EQTR loops. Qualcomm platforms would use
    this module parameter.
>
>> +/* A HS-G6 capable M-TX shall support the presets. */
>> +static const struct __ufs_tx_eq_preset {
>> +    unsigned int preshoot;
>> +    unsigned int deemphasis;
>> +} ufs_tx_eq_preset[UFS_TX_EQ_PRESET_MAX] = {
>> +    [UFS_TX_EQ_PRESET_P0] = {UFS_TX_HS_PRESHOOT_DB_0P0, 
>> UFS_TX_HS_DEEMPHASIS_DB_0P0},
>> +    [UFS_TX_EQ_PRESET_P1] = {UFS_TX_HS_PRESHOOT_DB_0P0, 
>> UFS_TX_HS_DEEMPHASIS_DB_0P8},
>> +    [UFS_TX_EQ_PRESET_P2] = {UFS_TX_HS_PRESHOOT_DB_0P0, 
>> UFS_TX_HS_DEEMPHASIS_DB_1P6},
>> +    [UFS_TX_EQ_PRESET_P3] = {UFS_TX_HS_PRESHOOT_DB_0P8, 
>> UFS_TX_HS_DEEMPHASIS_DB_0P0},
>> +    [UFS_TX_EQ_PRESET_P4] = {UFS_TX_HS_PRESHOOT_DB_1P6, 
>> UFS_TX_HS_DEEMPHASIS_DB_0P0},
>> +    [UFS_TX_EQ_PRESET_P5] = {UFS_TX_HS_PRESHOOT_DB_0P8, 
>> UFS_TX_HS_DEEMPHASIS_DB_0P8},
>> +    [UFS_TX_EQ_PRESET_P6] = {UFS_TX_HS_PRESHOOT_DB_0P8, 
>> UFS_TX_HS_DEEMPHASIS_DB_1P6},
>> +    [UFS_TX_EQ_PRESET_P7] = {UFS_TX_HS_PRESHOOT_DB_1P6, 
>> UFS_TX_HS_DEEMPHASIS_DB_0P8},
>> +};
>
> Please mention in the comment above this array from what standard the 
> above table comes.
Will do.
>
>> +static const u32 pa_peer_rx_adapt_initial[UFS_HS_GEAR_MAX] = {
>> +    0,
>> +    0,
>> +    0,
>> +    0,
>> +    PA_PEERRXHSG4ADAPTINITIAL,
>> +    PA_PEERRXHSG5ADAPTINITIAL,
>> +    PA_PEERRXHSG6ADAPTINITIALL0L3
>> +};
>> +
>> +static const u32 rx_adapt_initial_cap[UFS_HS_GEAR_MAX] = {
>> +    0,
>> +    0,
>> +    0,
>> +    0,
>> +    RX_HS_G4_ADAPT_INITIAL_CAP,
>> +    RX_HS_G5_ADAPT_INITIAL_CAP,
>> +    RX_HS_G6_ADAPT_INITIAL_CAP
>> +};
>> +
>> +static const u32 pa_tx_eq_setting[UFS_HS_GEAR_MAX] = {
>> +    0,
>> +    PA_TXEQG1SETTING,
>> +    PA_TXEQG2SETTING,
>> +    PA_TXEQG3SETTING,
>> +    PA_TXEQG4SETTING,
>> +    PA_TXEQG5SETTING,
>> +    PA_TXEQG6SETTING
>> +};
>
> Same comment for the above three arrays. Please add a comment that 
> explains what standard these arrays come from.
Sure.
>
>> +    if (!local_precodeen && !peer_precodeen) {
>> +        dev_dbg(hba->dev, "Pre-Coding is not required for either 
>> side\n");
>> +        return ret;
>> +    }
>
> Here and elsewhere in this patch, please change "precodeen" into 
> "precode_en". That will make it easier for readers to guess that
> "precode_en" refers to enabling precoding.
Sure.
>
>> +/**
>> + * ufshcd_evaluate_fom - Update TX EQ params based on FOM results
>> + * @hba: per adapter instance
>> + * @params: TX EQ parameters data structure
>> + * @h_iter: host TX EQTR iterator data structure
>> + * @d_iter: device TX EQTR iterator data structure
>> + *
>> + * Evaluate FOM results, update host and device TX EQ params if FOM 
>> results are
>> + * improved, and record TX EQTR results.
>> + */
>> +static void ufshcd_evaluate_fom(struct ufs_hba *hba,
>> +                struct ufshcd_tx_eq_params *params,
>> +                struct tx_eqtr_iter *h_iter,
>> +                struct tx_eqtr_iter *d_iter)
>> +{
>> +    u32 preshoot, deemphasis, fom_value;
>> +    bool precode_en;
>> +    int lane;
>> +
>> +    for (lane = 0; h_iter->is_new && lane < h_iter->num_lanes; 
>> lane++) {
>> +        preshoot = h_iter->preshoot;
>> +        deemphasis = h_iter->deemphasis;
>> +        fom_value = h_iter->fom[lane] & RX_FOM_VALUE_MASK;
>> +        precode_en = !!(h_iter->fom[lane] & RX_FOM_PRECODING_EN_MASK);
>
> !! is superfluous when assigning to a boolean (precode_en).
Will remove !!.
>
>> +static int __ufshcd_tx_eqtr(struct ufs_hba *hba,
>> +                struct ufshcd_tx_eq_params *params,
>> +                struct ufs_pa_layer_attr *pwr_mode)
>> +{
>> +    struct ufshcd_tx_eq_params *new_params;
>> +    struct tx_eqtr_iter h_iter, d_iter;
>> +    unsigned int preshoot, deemphasis;
>> +    u32 gear = pwr_mode->gear_tx;
>> +    ktime_t start;
>> +    int ret;
>> +
>> +    new_params = kzalloc(sizeof(struct ufshcd_tx_eq_params), 
>> GFP_KERNEL);
>> +    if (!new_params)
>> +        return -ENOMEM;
>
> Please combine the declaration of 'new_params' with its assignment and
> use __free to simplify error handling, e.g. as follows:
>
> struct ufshcd_tx_eq_params *new_params __free(kfree) = 
> kzalloc(sizeof(struct ufshcd_tx_eq_params), GFP_KERNEL);
Thanks for the suggestion.
>
>> +    /* TX EQTR main loop */
>> +    for (preshoot = 0; preshoot < TX_HS_NUM_PRESHOOT; preshoot++)
>
> Please surround the body of this loop with braces ({}).
Will do.
>
>> +        for (deemphasis = 0; deemphasis < TX_HS_NUM_DEEMPHASIS; 
>> deemphasis++) {
>> +            if (!tx_eqtr_iter_update(preshoot, deemphasis, &h_iter, 
>> &d_iter))
>> +                continue;
>> +
>> +            /* Step 3 - Apply TX EQTR settings */
>> +            ret = ufshcd_apply_tx_eqtr_settings(hba, pwr_mode, 
>> &h_iter, &d_iter);
>> +            if (ret) {
>> +                dev_err(hba->dev, "Failed to apply TX EQTR settings: 
>> %d\n",
>> +                    ret);
>> +                goto out;
>> +            }
>> +
>> +            /* Step 4 - Trigger TX EQTR procedure start */
>> +            ret = ufshcd_trigger_tx_eqtr(hba, gear);
>> +            if (ret) {
>> +                dev_err(hba->dev, "Failed to start TX EQTR procedure 
>> for target gear %u: %d\n",
>> +                    gear, ret);
>> +                goto out;
>> +            }
>> +
>> +            /* Step 5 - Get FOM */
>> +            ret = ufshcd_get_rx_fom(hba, pwr_mode, &h_iter, &d_iter);
>> +            if (ret) {
>> +                dev_err(hba->dev, "Failed to get RX_FOM: %d\n",
>> +                    ret);
>> +                goto out;
>> +            }
>> +
>> +            ufshcd_evaluate_fom(hba, new_params, &h_iter, &d_iter);
>> +    };
>
> Please remove the superfluous semicolon past }.
Sorry, my bad.
>
>> + * It ensures that EQTR starts from the most reiliable link state 
>> (HS-G1) with
>
> reiliable -> reliable
Thanks.
>
>> +static void ufshcd_tx_eqtr_unprepare(struct ufs_hba *hba,
>> +                     struct ufs_pa_layer_attr *pwr_mode)
>> +{
>> +    int err;
>> +
>> +    if (pwr_mode->pwr_rx == SLOWAUTO_MODE || pwr_mode->hs_rate == 0)
>> +        return;
>> +
>> +    err = ufshcd_change_power_mode(hba, pwr_mode, /*force_pmc=*/false);
>> +    if (err)
>> +        dev_err(hba->dev, "%s: Failed to restore Power Mode: %d\n",
>> +            __func__, err);
>> +}
>
> What should happen if restoring the power mode fails?
I made ufshcd_tx_eqtr_unprepare() void because I don't want power mode
restoring (if successful) to override the actual err which leads to power
mode restoring in the first place.

ufshcd_tx_eqtr(), caller of ufshcd_tx_eqtr_unprepare(), would anyways
return the actual err to its caller regardless of power mode restoring
succeeds or not. Eventually, the ufshcd_config_pwr_mode() would return
the same err to its caller.

Currently the two paths which call into ufshcd_config_pwr_mode() are
probe() and clock scaling.

For probe(), if ufshcd_config_pwr_mode() returns any err, probe() would
fail anyways regardless of if power mode is restored to HS-G1.

For clock scaling, if ufshcd_config_pwr_mode() returns err and
ufshcd_tx_eqtr_unprepare() couldn't restore to original power mode,
the worst case is that power mode is left as HS-G1 (the most reliable gear).

Do you see a problem with this approach?
>
>> +int ufshcd_config_tx_eq_settings(struct ufs_hba *hba,
>> +                 struct ufs_pa_layer_attr *pwr_mode)
>> +{
>> +    struct ufshcd_tx_eq_params *params;
>> +    u32 gear, rate;
>> +    int ret = 0;
>
> Please remove the local variable 'ret' and move it into the two scopes
> in which it is assigned a value.
OK
>
>> +    if (!ufshcd_is_tx_eq_supported(hba) || !use_adaptive_txeq)
>> +        return ret;
>
> To improve code readability, please change "return ret" into "return 0".
Will do.
>
>> +    } else if (gear < adaptive_txeq_gear) {
>> +        return ret;
>> +    }
>
> To improve code readability, please change "return ret" into "return 0".
Will do.
>
>> +        params->is_applied = true;
>> +    }
>> +
>> +    return ret;
>
> To improve code readability, please change "return ret" into "return 0".
Will do.
>
>>   /*
>>    * PHY Adapter attributes
>>    */
>> -#define PA_PHY_TYPE        0x1500
>
> Please refrain from making whitespace changes in existing code in a 
> patch that is already too big.
OK.
>
>>   /* Adpat type for PA_TXHSADAPTTYPE attribute */
>
> Adpat -> adapt
Thanks.

Best Regards,
Can Guo.
>
> Thanks,
>
> Bart.