Re: [PATCH v3 7/8] thermal: exynos: split initialization of TMU and the thermal zone

[Marek Szyprowski]

On 23.10.2023 14:59, Lukasz Luba wrote:
> On 10/3/23 12:16, Mateusz Majewski wrote:
>> This will be needed in the future, as the thermal zone subsystem might
>> call our callbacks right after devm_thermal_of_zone_register. Currently
>> we just make get_temp return EAGAIN in such case, but this will not be
>> possible with state-modifying callbacks, for instance set_trips.
>>
>> Signed-off-by: Mateusz Majewski <m.majewski2@xxxxxxxxxxx>
>> ---
>> v1 -> v2: We take clocks into account; exynos_tmu_initialize needs both
>>    clocks, as tmu_initialize might use the base_second registers. 
>> However,
>>    exynos_thermal_zone_configure only needs clk.
>>
>>   drivers/thermal/samsung/exynos_tmu.c | 104 +++++++++++++++------------
>>   1 file changed, 60 insertions(+), 44 deletions(-)
>>
>> diff --git a/drivers/thermal/samsung/exynos_tmu.c 
>> b/drivers/thermal/samsung/exynos_tmu.c
>> index 7138e001fa5a..343e27c61528 100644
>> --- a/drivers/thermal/samsung/exynos_tmu.c
>> +++ b/drivers/thermal/samsung/exynos_tmu.c
>> @@ -251,25 +251,8 @@ static void sanitize_temp_error(struct 
>> exynos_tmu_data *data, u32 trim_info)
>>   static int exynos_tmu_initialize(struct platform_device *pdev)
>>   {
>>       struct exynos_tmu_data *data = platform_get_drvdata(pdev);
>> -    struct thermal_zone_device *tzd = data->tzd;
>> -    int num_trips = thermal_zone_get_num_trips(tzd);
>>       unsigned int status;
>> -    int ret = 0, temp;
>> -
>> -    ret = thermal_zone_get_crit_temp(tzd, &temp);
>> -    if (ret && data->soc != SOC_ARCH_EXYNOS5433) { /* FIXME */
>> -        dev_err(&pdev->dev,
>> -            "No CRITICAL trip point defined in device tree!\n");
>> -        goto out;
>> -    }
>> -
>> -    if (num_trips > data->ntrip) {
>> -        dev_info(&pdev->dev,
>> -             "More trip points than supported by this TMU.\n");
>> -        dev_info(&pdev->dev,
>> -             "%d trip points should be configured in polling mode.\n",
>> -             num_trips - data->ntrip);
>> -    }
>> +    int ret = 0;
>>         mutex_lock(&data->lock);
>>       clk_enable(data->clk);
>> @@ -280,32 +263,63 @@ static int exynos_tmu_initialize(struct 
>> platform_device *pdev)
>>       if (!status) {
>>           ret = -EBUSY;
>>       } else {
>> -        int i, ntrips =
>> -            min_t(int, num_trips, data->ntrip);
>> -
>>           data->tmu_initialize(pdev);
>> -
>> -        /* Write temperature code for rising and falling threshold */
>> -        for (i = 0; i < ntrips; i++) {
>> -
>> -            struct thermal_trip trip;
>> -
>> -            ret = thermal_zone_get_trip(tzd, i, &trip);
>> -            if (ret)
>> -                goto err;
>> -
>> -            data->tmu_set_trip_temp(data, i, trip.temperature / 
>> MCELSIUS);
>> -            data->tmu_set_trip_hyst(data, i, trip.temperature / 
>> MCELSIUS,
>> -                        trip.hysteresis / MCELSIUS);
>> -        }
>> -
>>           data->tmu_clear_irqs(data);
>>       }
>> +
>> +    mutex_unlock(&data->lock);
>> +    clk_disable(data->clk);
>> +    if (!IS_ERR(data->clk_sec))
>> +        clk_disable(data->clk_sec);
>
> In all other places the clock is strictly protected inside the critical
> section, but not here. In this code in theory on SMP (especially with
> big.LITTLE system with different speeds of CPUs) this could lead to
> disabling the clocks just after other CPU acquired the mutex and enabled
> them (in order to use the HW regs).


Clocks have internal atomic counters, so it is safe to disable them 
after leaving critical section. The clock would still be enabled in the 
mentioned case.


> Please put those two clocks before the mutex_unlock() and in the
> reverse order.

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland