Message-ID: <22278de4-ade3-45fc-8cd5-dd35043ed6c8@bootlin.com>
Date: Wed, 24 Apr 2024 11:25:09 +0200
From: Thomas Richard <thomas.richard@...tlin.com>
To: Daniel Lezcano <daniel.lezcano@...aro.org>, rafael@...nel.org
Cc: rui.zhang@...el.com, lukasz.luba@....com, linux-pm@...r.kernel.org,
 linux-kernel@...r.kernel.org, gregory.clement@...tlin.com,
 thomas.petazzoni@...tlin.com, theo.lebrun@...tlin.com, u-kumar1@...com,
 Keerthy <j-keerthy@...com>
Subject: Re: [PATCH v3] thermal: k3_j72xx_bandgap: implement suspend/resume
 support

On 4/22/24 17:46, Daniel Lezcano wrote:
> On 17/04/2024 14:03, Thomas Richard wrote:
>> From: Théo Lebrun <theo.lebrun@...tlin.com>
>>
>> This add suspend-to-ram support.
>>
>> The derived_table is kept-as is, so the resume is only about
>> pm_runtime_* calls and restoring the same registers as the probe.
>>
>> Extract the hardware initialization procedure to a function called at
>> both probe-time & resume-time.
>>
>> The probe-time loop is split in two to ensure doing the hardware
>> initialization before registering thermal zones. That ensures our
>> callbacks cannot be called while in bad state.
>>
>> Signed-off-by: Théo Lebrun <theo.lebrun@...tlin.com>
>> Acked-by: Keerthy <j-keerthy@...com>
>> Signed-off-by: Thomas Richard <thomas.richard@...tlin.com>
>> ---
>>
>> v3:
>>   - Remove __maybe_unused attributes and use the magic of PTR_IF()
> 
> Where are these changes ?

- I removed __maybe_unused attribute for k3_j72xx_bandgap_suspend() and
k3_j72xx_bandgap_resume()
- The DEFINE_SIMPLE_DEV_PM_OPS macro is used to declare
k3_j72xx_bandgap_pm_ops.
- The pm_sleep_ptr macro is used.

> 
>> v2:
>>   - Fix warnings/errors reported by kernel test robot
>>
>>   drivers/thermal/k3_j72xx_bandgap.c | 112 ++++++++++++++++++++---------
>>   1 file changed, 79 insertions(+), 33 deletions(-)
>>
>> diff --git a/drivers/thermal/k3_j72xx_bandgap.c
>> b/drivers/thermal/k3_j72xx_bandgap.c
>> index c74094a86982..a7244e05ed5c 100644
>> --- a/drivers/thermal/k3_j72xx_bandgap.c
>> +++ b/drivers/thermal/k3_j72xx_bandgap.c
>> @@ -178,6 +178,7 @@ struct k3_j72xx_bandgap {
>>       void __iomem *base;
>>       void __iomem *cfg2_base;
>>       struct k3_thermal_data *ts_data[K3_VTM_MAX_NUM_TS];
>> +    int cnt;
>>   };
>>     /* common data structures */
>> @@ -338,24 +339,53 @@ static void print_look_up_table(struct device
>> *dev, int *ref_table)
>>           dev_dbg(dev, "%d       %d %d\n", i, derived_table[i],
>> ref_table[i]);
>>   }
>>   +static void k3_j72xx_bandgap_init_hw(struct k3_j72xx_bandgap *bgp)
>> +{
>> +    struct k3_thermal_data *data;
>> +    int id, high_max, low_temp;
>> +    u32 val;
>> +
>> +    for (id = 0; id < bgp->cnt; id++) {
>> +        data = bgp->ts_data[id];
>> +        val = readl(bgp->cfg2_base + data->ctrl_offset);
>> +        val |= (K3_VTM_TMPSENS_CTRL_MAXT_OUTRG_EN |
>> +            K3_VTM_TMPSENS_CTRL_SOC |
>> +            K3_VTM_TMPSENS_CTRL_CLRZ | BIT(4));
>> +        writel(val, bgp->cfg2_base + data->ctrl_offset);
>> +    }
>> +
>> +    /*
>> +     * Program TSHUT thresholds
>> +     * Step 1: set the thresholds to ~123C and 105C WKUP_VTM_MISC_CTRL2
>> +     * Step 2: WKUP_VTM_TMPSENS_CTRL_j set the MAXT_OUTRG_EN  bit
>> +     *         This is already taken care as per of init
>> +     * Step 3: WKUP_VTM_MISC_CTRL set the ANYMAXT_OUTRG_ALERT_EN  bit
>> +     */
>> +    high_max = k3_j72xx_bandgap_temp_to_adc_code(MAX_TEMP);
>> +    low_temp = k3_j72xx_bandgap_temp_to_adc_code(COOL_DOWN_TEMP);
>> +
>> +    writel((low_temp << 16) | high_max, bgp->cfg2_base +
>> K3_VTM_MISC_CTRL2_OFFSET);
>> +    mdelay(100);
> 
>     100ms ?
> 
> Ok, I know you did not introduce this delay. But as it is a suspend /
> resume, this delay will be called at each resume which is not acceptable
> given its duration.
> 
> Please, investigate a way to get rid of this delay

I talked to Keerthy, this delay was initially added to be sure the
thresholds are programmed before enabling the interrupt.
But for him, this delay is useless.

I did the test, I removed the delay and I didn't see any thermal events.
So we can remove it.
I'll send a new version without the delay, and I'll mention it in the
commit message.

> 
> 
>> +    writel(K3_VTM_ANYMAXT_OUTRG_ALERT_EN, bgp->cfg2_base +
>> K3_VTM_MISC_CTRL_OFFSET);
>> +}
>> +
>>   struct k3_j72xx_bandgap_data {
>>       const bool has_errata_i2128;
>>   };
>>     static int k3_j72xx_bandgap_probe(struct platform_device *pdev)
>>   {
>> -    int ret = 0, cnt, val, id;
>> -    int high_max, low_temp;
>> -    struct resource *res;
>> +    const struct k3_j72xx_bandgap_data *driver_data;
>> +    struct thermal_zone_device *ti_thermal;
>>       struct device *dev = &pdev->dev;
>> +    bool workaround_needed = false;
>>       struct k3_j72xx_bandgap *bgp;
>>       struct k3_thermal_data *data;
>> -    bool workaround_needed = false;
>> -    const struct k3_j72xx_bandgap_data *driver_data;
>> -    struct thermal_zone_device *ti_thermal;
>> -    int *ref_table;
>>       struct err_values err_vals;
>>       void __iomem *fuse_base;
>> +    int ret = 0, val, id;
>> +    struct resource *res;
>> +    int *ref_table;
> 
> No related to the changes but the patch is showing signs the driver
> needs a code cleanup AFAICT
> 
>>       const s64 golden_factors[] = {
>>           -490019999999999936,
>> @@ -422,10 +452,10 @@ static int k3_j72xx_bandgap_probe(struct
>> platform_device *pdev)
>>         /* Get the sensor count in the VTM */
>>       val = readl(bgp->base + K3_VTM_DEVINFO_PWR0_OFFSET);
>> -    cnt = val & K3_VTM_DEVINFO_PWR0_TEMPSENS_CT_MASK;
>> -    cnt >>= __ffs(K3_VTM_DEVINFO_PWR0_TEMPSENS_CT_MASK);
>> +    bgp->cnt = val & K3_VTM_DEVINFO_PWR0_TEMPSENS_CT_MASK;
>> +    bgp->cnt >>= __ffs(K3_VTM_DEVINFO_PWR0_TEMPSENS_CT_MASK);
>>   -    data = devm_kcalloc(bgp->dev, cnt, sizeof(*data), GFP_KERNEL);
>> +    data = devm_kcalloc(bgp->dev, bgp->cnt, sizeof(*data), GFP_KERNEL);
>>       if (!data) {
>>           ret = -ENOMEM;
>>           goto err_alloc;
>> @@ -449,8 +479,8 @@ static int k3_j72xx_bandgap_probe(struct
>> platform_device *pdev)
>>       else
>>           init_table(3, ref_table, pvt_wa_factors);
>>   -    /* Register the thermal sensors */
>> -    for (id = 0; id < cnt; id++) {
>> +    /* Precompute the derived table & fill each thermal sensor struct */
>> +    for (id = 0; id < bgp->cnt; id++) {
>>           data[id].bgp = bgp;
>>           data[id].ctrl_offset = K3_VTM_TMPSENS0_CTRL_OFFSET + id * 0x20;
>>           data[id].stat_offset = data[id].ctrl_offset +
>> @@ -470,13 +500,13 @@ static int k3_j72xx_bandgap_probe(struct
>> platform_device *pdev)
>>           else if (id == 0 && !workaround_needed)
>>               memcpy(derived_table, ref_table, TABLE_SIZE * 4);
>>   -        val = readl(data[id].bgp->cfg2_base + data[id].ctrl_offset);
>> -        val |= (K3_VTM_TMPSENS_CTRL_MAXT_OUTRG_EN |
>> -            K3_VTM_TMPSENS_CTRL_SOC |
>> -            K3_VTM_TMPSENS_CTRL_CLRZ | BIT(4));
>> -        writel(val, data[id].bgp->cfg2_base + data[id].ctrl_offset);
>> -
>>           bgp->ts_data[id] = &data[id];
>> +    }
>> +
>> +    k3_j72xx_bandgap_init_hw(bgp);
>> +
>> +    /* Register the thermal sensors */
>> +    for (id = 0; id < bgp->cnt; id++) {
>>           ti_thermal = devm_thermal_of_zone_register(bgp->dev, id,
>> &data[id],
>>                                  &k3_of_thermal_ops);
>>           if (IS_ERR(ti_thermal)) {
>> @@ -486,21 +516,7 @@ static int k3_j72xx_bandgap_probe(struct
>> platform_device *pdev)
>>           }
>>       }
>>   -    /*
>> -     * Program TSHUT thresholds
>> -     * Step 1: set the thresholds to ~123C and 105C WKUP_VTM_MISC_CTRL2
>> -     * Step 2: WKUP_VTM_TMPSENS_CTRL_j set the MAXT_OUTRG_EN  bit
>> -     *         This is already taken care as per of init
>> -     * Step 3: WKUP_VTM_MISC_CTRL set the ANYMAXT_OUTRG_ALERT_EN  bit
>> -     */
>> -    high_max = k3_j72xx_bandgap_temp_to_adc_code(MAX_TEMP);
>> -    low_temp = k3_j72xx_bandgap_temp_to_adc_code(COOL_DOWN_TEMP);
>> -
>> -    writel((low_temp << 16) | high_max, data[0].bgp->cfg2_base +
>> -           K3_VTM_MISC_CTRL2_OFFSET);
>> -    mdelay(100);
>> -    writel(K3_VTM_ANYMAXT_OUTRG_ALERT_EN, data[0].bgp->cfg2_base +
>> -           K3_VTM_MISC_CTRL_OFFSET);
>> +    platform_set_drvdata(pdev, bgp);
>>         print_look_up_table(dev, ref_table);
>>       /*
>> @@ -527,6 +543,35 @@ static void k3_j72xx_bandgap_remove(struct
>> platform_device *pdev)
>>       pm_runtime_disable(&pdev->dev);
>>   }
>>   +static int k3_j72xx_bandgap_suspend(struct device *dev)
>> +{
>> +    pm_runtime_put_sync(dev);
>> +    pm_runtime_disable(dev);
>> +    return 0;
>> +}
>> +
>> +static int k3_j72xx_bandgap_resume(struct device *dev)
>> +{
>> +    struct k3_j72xx_bandgap *bgp = dev_get_drvdata(dev);
>> +    int ret;
>> +
>> +    pm_runtime_enable(dev);
>> +    ret = pm_runtime_get_sync(dev);
>> +    if (ret < 0) {
>> +        pm_runtime_put_noidle(dev);
>> +        pm_runtime_disable(dev);
>> +        return ret;
>> +    }
>> +
>> +    k3_j72xx_bandgap_init_hw(bgp);
>> +
>> +    return 0;
>> +}
>> +
>> +static DEFINE_SIMPLE_DEV_PM_OPS(k3_j72xx_bandgap_pm_ops,
>> +                k3_j72xx_bandgap_suspend,
>> +                k3_j72xx_bandgap_resume);
>> +
>>   static const struct k3_j72xx_bandgap_data
>> k3_j72xx_bandgap_j721e_data = {
>>       .has_errata_i2128 = true,
>>   };
>> @@ -554,6 +599,7 @@ static struct platform_driver
>> k3_j72xx_bandgap_sensor_driver = {
>>       .driver = {
>>           .name = "k3-j72xx-soc-thermal",
>>           .of_match_table    = of_k3_j72xx_bandgap_match,
>> +        .pm = pm_sleep_ptr(&k3_j72xx_bandgap_pm_ops),
>>       },
>>   };
>>   
> 
-- 
Thomas Richard, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

Powered by blists - more mailing lists