Message-ID: <0ac96eb0-908c-f0c6-2ec8-23957f5a7941@amd.com>
Date:   Fri, 24 Mar 2023 14:26:38 -0700
From:   Lizhi Hou <lizhi.hou@....com>
To:     Rob Herring <robh@...nel.org>
CC:     <linux-pci@...r.kernel.org>, <devicetree@...r.kernel.org>,
        <linux-kernel@...r.kernel.org>, <frowand.list@...il.com>,
        <helgaas@...nel.org>, <clement.leger@...tlin.com>,
        <max.zhen@....com>, <sonal.santan@....com>, <larry.liu@....com>,
        <brian.xu@....com>, <stefano.stabellini@...inx.com>,
        <trix@...hat.com>
Subject: Re: [PATCH V7 1/3] of: dynamic: Add interfaces for creating device
 node dynamically


On 3/24/23 07:14, Rob Herring wrote:
> On Thu, Mar 23, 2023 at 9:12 PM Lizhi Hou <lizhi.hou@....com> wrote:
>>
>> On 3/23/23 15:40, Rob Herring wrote:
>>> On Thu, Jan 19, 2023 at 9:02 PM Lizhi Hou <lizhi.hou@....com> wrote:
>>>> of_create_node() creates device node dynamically. The parent device node
>>>> and full name are required for creating the node. It optionally creates
>>>> an OF changeset and attaches the newly created node to the changeset. The
>>>> device node pointer and the changeset pointer can be used to add
>>>> properties to the device node and apply the node to the base tree.
>>>>
>>>> of_destroy_node() frees the device node created by of_create_node(). If
>>>> an OF changeset was also created for this node, it will destroy the
>>>> changeset before freeing the device node.
>>>>
>>>> Expand of_changeset APIs to handle specific types of properties.
>>>>       of_changeset_add_prop_string()
>>>>       of_changeset_add_prop_string_array()
>>>>       of_changeset_add_prop_u32_array()
>>>>
>>>> Signed-off-by: Lizhi Hou <lizhi.hou@....com>
>>> Your Sob should be last because you sent this patch. The order of Sob
>>> is roughly the order of possession of the patch.
>> Got it.
>>>> Signed-off-by: Sonal Santan <sonal.santan@....com>
>>>> Signed-off-by: Max Zhen <max.zhen@....com>
>>> So Sonal and Max modified this patch?
>> They did not directly modify the code. And we discussed the design
>> together.  They also reviewed the patch before I sent it out. Please let
>> me know if other keyword should be used in this case.
> Reviewed-by or nothing. Some feel that only reviews on public lists
> should get that tag and internal, private reviews don't matter.
>
>>>> Reviewed-by: Brian Xu <brian.xu@....com>
>>>> Signed-off-by: Clément Léger <clement.leger@...tlin.com>
>>> Why does this have Clément's Sob?
>> I referenced Clément 's code and used one portion in my first patch
>> series. And I re-implemented it later to address the code review
>> comments/requests.
> Then it goes first or you can use the 'Co-developed-by' tag.
>
>>>> ---
>>>>    drivers/of/dynamic.c | 197 +++++++++++++++++++++++++++++++++++++++++++
>>>>    include/linux/of.h   |  24 ++++++
>>>>    2 files changed, 221 insertions(+)
>>>>
>>>> diff --git a/drivers/of/dynamic.c b/drivers/of/dynamic.c
>>>> index cd3821a6444f..4e211a1d039f 100644
>>>> --- a/drivers/of/dynamic.c
>>>> +++ b/drivers/of/dynamic.c
>>>> @@ -461,6 +461,71 @@ struct device_node *__of_node_dup(const struct device_node *np,
>>>>           return NULL;
>>>>    }
>>>>
>>>> +/**
>>>> + * of_create_node - Dynamically create a device node
>>> For consistency, I think this should be of_changeset_create_node().
>> Sure.
>>>> + *
>>>> + * @parent: Pointer to parent device node
>>>> + * @full_name: Node full name
>>>> + * @cset: Pointer to returning changeset
>>>> + *
>>>> + * Return: Pointer to the created device node or NULL in case of an error.
>>>> + */
>>>> +struct device_node *of_create_node(struct device_node *parent,
>>>> +                                  const char *full_name,
>>>> +                                  struct of_changeset **cset)
>>>> +{
>>>> +       struct of_changeset *ocs;
>>>> +       struct device_node *np;
>>>> +       int ret;
>>>> +
>>>> +       np = __of_node_dup(NULL, full_name);
>>>> +       if (!np)
>>>> +               return NULL;
>>>> +       np->parent = parent;
>>>> +
>>>> +       if (!cset)
>>>> +               return np;
>>>> +
>>>> +       ocs = kmalloc(sizeof(*ocs), GFP_KERNEL);
>>>> +       if (!ocs) {
>>>> +               of_node_put(np);
>>>> +               return NULL;
>>>> +       }
>>>> +
>>>> +       of_changeset_init(ocs);
>>>> +       ret = of_changeset_attach_node(ocs, np);
>>>> +       if (ret) {
>>>> +               of_changeset_destroy(ocs);
>>>> +               of_node_put(np);
>>>> +               kfree(ocs);
>>>> +               return NULL;
>>>> +       }
>>>> +
>>>> +       np->data = ocs;
>>>> +       *cset = ocs;
>>>> +
>>>> +       return np;
>>>> +}
>>>> +EXPORT_SYMBOL(of_create_node);
>>>> +
>>>> +/**
>>>> + * of_destroy_node - Destroy a dynamically created device node
>>>> + *
>>>> + * @np: Pointer to dynamically created device node
>>>> + *
>>>> + */
>>>> +void of_destroy_node(struct device_node *np)
>>>> +{
>>>> +       struct of_changeset *ocs;
>>>> +
>>>> +       if (np->data) {
>>>> +               ocs = (struct of_changeset *)np->data;
>>>> +               of_changeset_destroy(ocs);
>>>> +       }
>>>> +       of_node_put(np);
>>> A sequence like this would be broken:
>>>
>>> np  = of_create_node()
>>> of_node_get(np)
>>> of_destroy_node(np)
>>>
>>> The put here won't free the node because it still has a ref, but we
>>> just freed the changeset. For this to work correctly, we would need
>>> the release function to handle np->data instead. However, all users of
>>> data aren't a changeset.
>>>
>>> I'm failing to remember why we're storing the changeset in 'data', but
>>> there doesn't seem to be a reason now so I think that can just be
>>> dropped. Then if you want to free the node, you'd just do an
>>> of_node_put(). (And maybe after the node is attached you do a put too,
>>> because the attach does a get. Not completely sure.)
>> The question is how to save changeset and free it later. I used global
>> link list to track the changeset been created.
>>
>> Storing the changeset in 'data' can avoid using the global link list.
>>
>> To use of_node_put() to free both node and changeset, I think we can
>>
>>     1) add a new flag, then in of_node_release() we can know np->data is
>> changeset by checking the flag.
>>
>>     2) When creating node, allocate extra memory for changeset and set
>> np->data to a global function of_free_dynamic_node().
>>
>>         In of_node_release(), check if np->data == of_free_dynamic_node,
>> call of_free_dynamic_node(np).
>>
>>         in of_free_dynamic_node(), free changeset by
>> of_changeset_destroy(np+1)
>>
>> Does this make sense to you? If yes, 1) or 2) sounds better?
> Neither works. Changesets and nodes are not 1:1 in general though they
> are in your use. So you can use the data ptr, but the caller has to
> decide that, not the DT core code.

Ok. In of_pci_make_dev_node(), I can do

      ocs = kmalloc(*ocs);

      of_changeset_init(ocs);

      np = of_changeset_create_node(ocs, name);

      np->data = ocs;

Then in of_pci_remove_node(), I can do

      if (!np || !of_node_check_flag(np, OF_DYNAMIC)) return;

     of_changeset_destroy(np->data);

     of_node_put(np);


Does this sound reasonable?


Thanks,

Lizhi

>
> Rob

Powered by blists - more mailing lists