On 1/6/25 9:33 AM, Tudor Ambarus wrote:
> Hi, Krzysztof,
>
> On 12/31/24 2:32 PM, Tudor Ambarus wrote:
>>>> diff --git a/drivers/firmware/samsung/Kconfig b/drivers/firmware/samsung/Kconfig
>>>> new file mode 100644
>>>> index 000000000000..750b41342174
>>>> --- /dev/null
>>>> +++ b/drivers/firmware/samsung/Kconfig
>>>> @@ -0,0 +1,14 @@
>>>> +# SPDX-License-Identifier: GPL-2.0-only
>>>> +
>>>> +config EXYNOS_ACPM_PROTOCOL
>>>> + tristate "Exynos Alive Clock and Power Manager (ACPM) Message Protocol"
>>>> + depends on ARCH_EXYNOS || COMPILE_TEST
>>>> + depends on EXYNOS_MBOX
>>>
>>> Is it build time dependency? No || COMPILE_TEST?
>>
>> There's no build time dependency, I'll drop this line.
>>>
>>> Is it fine when EXYNOS_MBOX is a module?
>>
>> Yes. When the EXYNOS_MBOX module is not loaded, and one tries to load
>> EXYNOS_ACPM_PROTOCOL module, the later will defer probe when requesting
>> the mailbox channels, but that's fine.
>>
>
> I'll need to select EXYNOS_MBOX, I explain why below.
>
> cut
>
>>>> + */
>>>> +static const struct acpm_handle *acpm_get_by_phandle(struct device_node *np,
>>>> + const char *property)
>>>> +{
>>>> + struct acpm_handle *handle = NULL;
>>>> + struct device_node *acpm_np;
>>>> + struct acpm_info *info;
>>>> +
>>>> + if (!np) {
>>>> + pr_err("I need a device pointer\n");
>>>> + return ERR_PTR(-EINVAL);
>>>> + }
>>>> +
>>>> + acpm_np = of_parse_phandle(np, property, 0);
>>>> + if (!acpm_np)
>>>> + return ERR_PTR(-ENODEV);
>>>> +
>>>> + mutex_lock(&acpm_list_mutex);
>>>> + list_for_each_entry(info, &acpm_list, node) {
>>>> + if (acpm_np == info->dev->of_node) {
>>>> + handle = &info->handle;
>>>> + info->users++;
>>>> + break;
>>>> + }
>>>> + }
>>>> + mutex_unlock(&acpm_list_mutex);
>>>> + of_node_put(acpm_np);
>>>> +
>>>
>>> You also need device links and probably try_module_get. See clk.c
>
> I find these necessary too, will add them. try_module_get() must be
> called when the module exists and is alive, otherwise I get a NULL ptr
> dereference. I need a module dependency between acpm-protocol.ko and
> exynos-mailbox.ko.
> select EXYNOS_MBOX and
> MODULE_SOFTDEP("pre: exynos-mailbox");
> shall do the trick I think.
>
>>> clk_hw_create_clk() or of_qcom_ice_get(). Interestingly, none of them
>>> perform both operations, which I think is necessary.
>>>
>>> I think you could also avoid entire list and mutex by using
>>> platform_get_drvdata(), see of_qcom_ice_get().
>
> Using platform_get_drvdata() will simplify the code, thanks. It still
> assumes the platform driver exists and is alive, otherwise we get a NULL
> ptr dereference when getting the drvdata. But we'll be safe if I add the
> module dependency.
Ah, MODULE_SOFTDEP is a soft dependency, so I can't use
platform_get_drvdata(), because if someone removes the exynos-mailbox.ko
from the file system for example, the acpm protocol will defer and its
clients still get a NULL ptr dereference when trying to get the handle
(where try_module_get() and platform_get_drvdata() are called). I'm
better off with the list and mutex.
