Re: [PATCH v3 2/2] cpufreq: tegra: Re-model Tegra cpufreq driver

[Stephen Warren <swarren@wwwdotorg.org>]

On 12/09/2013 01:44 AM, bilhuang wrote:
> On 12/06/2013 07:04 AM, Stephen Warren wrote:
>> On 12/05/2013 12:44 AM, Bill Huang wrote:
>>> Re-model Tegra cpufreq driver to support all Tegra series of SoCs.
>>>
>>> * Make tegra-cpufreq.c a generic Tegra cpufreq driver.
>>> * Move Tegra20 specific codes into tegra20-cpufreq.c.
>>> * Bind Tegra cpufreq dirver with a fake device so defer probe would work
>>>    when we're going to get regulator in the driver to support voltage
>>>    scaling (DVFS).
>>
>>> diff --git a/drivers/cpufreq/tegra-cpufreq.c
>>> b/drivers/cpufreq/tegra-cpufreq.c
>>
>>> @@ -91,14 +40,10 @@ static int tegra_update_cpu_speed(struct
>>> cpufreq_policy *policy,
>> ...
>>> +    if (soc_config->vote_emc_on_cpu_rate)
>>> +        soc_config->vote_emc_on_cpu_rate(rate);
>>> +
>>> +    ret = soc_config->cpu_clk_set_rate(rate * 1000);
>>>       if (ret)
>>>           pr_err("cpu-tegra: Failed to set cpu frequency to %lu kHz\n",
>>>               rate);
>>
>> Is there any/much shared code left in this file after this patch? It
>> seems like all this file does now is make each cpufreq callback function
>> call soc_config->the_same_function_name(). If so, wouldn't it be better
>> to simply implement completely separate tegar20-cpufreq and
>> tegra30-cpufreq drivers, and register them each directly with the
>> cpufreq core, to avoid this file doing all the indirection?
>
> I think this file is needed since we can shared the registration and
> probe logic for different SoCs.

But there's basically nothing in probe() already, and if we have a
separate driver for each SoC, then there's even less code; just a call
to devm_kzalloc() for the device-specific data (which will be
SoC-specific in size anyway), and a call to cpufreq_register_driver(). I
don't think it's worth sharing that if it means that every other
function needs to be an indirect function call.

>>> -int __init tegra_cpufreq_init(void)
>>> +static struct {
>>> +    char *compat;
>>> +    int (*init)(struct tegra_cpufreq_data *,
>>> +            const struct tegra_cpufreq_config **);
>>> +} tegra_init_funcs[] = {
>>> +    { "nvidia,tegra20", tegra20_cpufreq_init },
>>> +};
>>> +
>>> +static int tegra_cpufreq_probe(struct platform_device *pdev)
>> ...
>>> +    for (i = 0; i < ARRAY_SIZE(tegra_init_funcs); i++) {
>>> +        if (of_machine_is_compatible(tegra_init_funcs[i].compat)) {
>>> +            ret = tegra_init_funcs[i].init(tegra_data, &soc_config);
>>> +            if (!ret)
>>> +                break;
>>> +            else
>>> +                goto out;
>>> +        }
>>>       }
>>> +    if (i == ARRAY_SIZE(tegra_init_funcs))
>>> +        goto out;
>>
>> I think there are better ways of doing this than open-coding it. Perhaps
>> of_match_device() or the platform-driver equivalent could be made to
>> work?
>
> Open coding is everywhere in OF helper functions actually. I doubt if we
> can use of_match_device() if we're not adding node in DT.
> If we're matching the platform device then we might need open coding, no?

For platform devices, you can set up the id_table of struct
platform_driver, and then simply call platform_get_device_id(pdev)
inside probe() to find the matching entry. drivers/i2c/busses/i2c-at91.c
is an example of how this works (just some random driver I found using
grep).

>>> +int __init tegra_cpufreq_init(void)
>>> +{
>>> +    struct platform_device_info devinfo = { .name = "tegra-cpufreq", };
>>> +
>>> +    platform_device_register_full(&devinfo);
>>> +
>>> +    return 0;
>>>   }
>>>   EXPORT_SYMBOL(tegra_cpufreq_init);
>>
>> Perhaps instead of hard-coding the name "tegra-cpufreq" here, you could
>> dynamically construct the device name based on the DT's root compatible
>> value, register "${root_compatible}-cpufreq", e.g.
>> "nvidia,tegra20-cpufreq" or "nvidia,tegra30-cpufreq". That would allow
>> the kernel's standard device/driver matching mechanism to pick the
>> correct driver to instantiate. Perhaps you could even dynamically
>> register an OF device so that you can use of_match_device() in probe, if
>
> I guess what you meant dynamically register an OF device is registering
> an fake OF device by calling of_device_add(), no? If yes then what
> of_node should we give?

Yes. Good question about which node. I guess the root node would be the
only one that made any sense at all, and admittedly it doesn't make a
huge amount of sense. Perhaps registers a platform device rather than an
OF device would make more sense. See platform_device_register() I think.