On 1/6/25 10:20 AM, Tudor Ambarus wrote:
>>>>> +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
>>>> 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().
[snip, irrelevant now]
I made my mind, I think the solution is threefold.
1/ use kref with a release callback so that the firmware drvdata will be
destroyed only when there are no consumers. Example:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=1e7c57355a3bc617fc220234889e49fe722a6305
2/ In get_handle() after kref_get_unless_zero() succeeds, I'll be
calling try_module_get() for supplier. Even if kref will make sure that
consumers will still have a valid supplier drvdata, I'd like to delete
the supplier module when there are no consumers left.
3/ use device_link_add() so that when the supplier device unbinds, to
unbind the consumer devices as well.
I'll implement these. Cheers,
ta
