Re: [PATCH v2] rpmsg: core: Add wildcard match for name service

[Arnaud POULIQUEN <arnaud.pouliquen@st.com>]

Hi Suman,

On 4/8/20 10:52 PM, Suman Anna wrote:
> On 4/8/20 10:59 AM, Mathieu Poirier wrote:
>> On Tue, 7 Apr 2020 at 17:07, Suman Anna <s-anna@ti.com> wrote:
>>>
>>> Hi Mathieu, Arnaud,
>>>
>>> On 3/27/20 2:36 PM, Mathieu Poirier wrote:
>>>> On Fri, Mar 27, 2020 at 10:35:34AM +0100, Arnaud POULIQUEN wrote:
>>>>> Hi
>>>>>
>>>>> On 3/26/20 11:01 PM, Mathieu Poirier wrote:
>>>>>> On Thu, 26 Mar 2020 at 14:42, Suman Anna <s-anna@ti.com> wrote:
>>>>>>>
>>>>>>> On 3/26/20 3:21 PM, Mathieu Poirier wrote:
>>>>>>>> On Thu, 26 Mar 2020 at 09:06, Suman Anna <s-anna@ti.com> wrote:
>>>>>>>>>
>>>>>>>>> Hi Mathieu,
>>>>>>>>>
>>>>>>>>> On 3/10/20 10:50 AM, Mathieu Poirier wrote:
>>>>>>>>>> Adding the capability to supplement the base definition published
>>>>>>>>>> by an rpmsg_driver with a postfix description so that it is possible
>>>>>>>>>> for several entity to use the same service.
>>>>>>>>>>
>>>>>>>>>> Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
>>>>>>>>>> Acked-by: Arnaud Pouliquen <arnaud.pouliquen@st.com>
>>>>>>>>>
>>>>>>>>> So, the concern I have here is that we are retrofitting this into the
>>>>>>>>> existing 32-byte name field, and the question is if it is going to be
>>>>>>>>> enough in general. That's the reason I went with the additional 32-byte
>>>>>>>>> field with the "rpmsg: add a description field" patch.
>>>>>>>>>
>>>>>>>>
>>>>>>>> That's a valid concern.
>>>>>>>>
>>>>>>>> Did you consider increasing the size of RPMSG_NAME_SIZE to 64? Have
>>>>>>>> you found cases where that wouldn't work?  I did a survey of all the
>>>>>>>> places the #define is used and all destination buffers are also using
>>>>>>>> the same #define in their definition.  It would also be backward
>>>>>>>> compatible with firmware implementations that use 32 byte.
>>>>>>>
>>>>>>> You can't directly bump the size without breaking the compatibility on
>>>>>>> the existing rpmsg_ns_msg in firmwares right? All the Linux-side drivers
>>>>>>> will be ok since they use the same macro but rpmsg_ns_msg has presence
>>>>>>> on both kernel and firmware-sides.
>>>>>>
>>>>>> Ah yes yes... The amount of bytes coming out of the pipe won't match.
>>>>>> Let me think a little...
>>>>>
>>>>> +1 for Suman's concern.
>>>>>
>>>>> Anyway i would like to challenge the need of more than 32 bytes to
>>>>> differentiate service instances.
>>>>> "AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA", seems to me enough if we only need
>>>>> to differentiate the instances.
>>>
>>> Remember that the rpmsg_device_id name takes some space within here. So,
>>> the shorter the rpmsg_device_id table name, the more room you have.
>>>
>>>>>
>>>>> But perhaps the need is also to provide a short description of the service?
>>>
>>> I am mostly using it to provide a unique instantiation name. In anycase,
>>> I have cross-checked against my current firmwares, and so far all of
>>> them happen to have the name + desc < 31 bytes.
>>>
>>>
>>>>>
>>>>> Suman, could you share some examples of your need?
>>>>
>>>> Looking at things further it is possible to extend the name of the service to
>>>> 64 byte while keeping backward compatibility by looking up the size of @len
>>>> in function rpmsg_ns_cb().  From there work with an rpmsg_ns_msg or a new
>>>> rpmsg_ns_msg64, pretty much the way you did in your patch[1].  In fact the
>>>> approach is the same except you are using 2 arrays of 32 byte and I'm using one
>>>> of 64.
>>>>
>>>> As Arnaud mentioned, is there an immediate need to support a 64-byte name?  If
>>>> not than I suggest to move forward with this patch and address the issue when we
>>>> get there - at least we know there is room for extention. Otherwise I'll spin
>>>> off another revision but it will be bigger and more complex.
>>>
>>> Yeah ok. I have managed to get my downstream drivers that use the desc
>>> field working with this patch after modifying the firmwares to publish
>>> using combined name, and adding logic in probe to get the trailing
>>> portion of the name.
>>
>> Perfect
>>
>>>
>>> So, the only thing that is missing or content for another patch is if we
>>> need to add some tooling/helper stuff for giving the trailing stuff to
>>> rpmsg drivers?
>>
>> So that all rpmsg drivers don't come up with their own parsing that
>> ends up doing the same thing.  Let me think about that - I may have to
>> get back to you...
> 
> Yep. Sure no problem. It can be a patch on top of this as well.
> 
> Arnaud,
> Do you have immediate need for the tooling stuff for the rpmsg-tty driver?

Before moving forward on rpmsg_tty i would prefer that we are aligned with Bjorn
on the implementation of the rpmsg_tty itself and the evolution of the service name...
Then rpmsg_tty could be a good threadfor a first implementation...


Concerning the name service, having a discussion around the name service skeleton would
be nice. This could be an good input for the helpers function definition.

Propositions in rpmsg_tty thread are

<service_name>-<feature>
or
<service_name>-<feature>-<sub_service>

don't hesitate to comment and/or propose alternatives

Regards,
Arnaud

> 
> regards
> Suman
> 
> 
>>
>>>
>>> regards
>>> Suman
>>>
>>>>
>>>> Thanks,
>>>> Mathieu
>>>>
>>>> [1]. https://patchwork.kernel.org/patch/11096599/
>>>>
>>>>>>>>>
>>>>>>>>>> ---
>>>>>>>>>> Changes for V2:
>>>>>>>>>> - Added Arnaud's Acked-by.
>>>>>>>>>> - Rebased to latest rproc-next.
>>>>>>>>>>
>>>>>>>>>>  drivers/rpmsg/rpmsg_core.c | 20 +++++++++++++++++++-
>>>>>>>>>>  1 file changed, 19 insertions(+), 1 deletion(-)
>>>>>>>>>>
>>>>>>>>>> diff --git a/drivers/rpmsg/rpmsg_core.c b/drivers/rpmsg/rpmsg_core.c
>>>>>>>>>> index e330ec4dfc33..bfd25978fa35 100644
>>>>>>>>>> --- a/drivers/rpmsg/rpmsg_core.c
>>>>>>>>>> +++ b/drivers/rpmsg/rpmsg_core.c
>>>>>>>>>> @@ -399,7 +399,25 @@ ATTRIBUTE_GROUPS(rpmsg_dev);
>>>>>>>>>>  static inline int rpmsg_id_match(const struct rpmsg_device *rpdev,
>>>>>>>>>>                                 const struct rpmsg_device_id *id)
>>>>>>>>>>  {
>>>>>>>>>> -     return strncmp(id->name, rpdev->id.name, RPMSG_NAME_SIZE) == 0;
>>>>>>>>>> +     size_t len = min_t(size_t, strlen(id->name), RPMSG_NAME_SIZE);
>>>>>>>>>> +
>>>>>>>>>> +     /*
>>>>>>>>>> +      * Allow for wildcard matches.  For example if rpmsg_driver::id_table
>>>>>>>>>> +      * is:
>>>>>>>>>> +      *
>>>>>>>>>> +      * static struct rpmsg_device_id rpmsg_driver_sample_id_table[] = {
>>>>>>>>>> +      *      { .name = "rpmsg-client-sample" },
>>>>>>>>>> +      *      { },
>>>>>>>>>> +      * }
>>>>>>>>>> +      *
>>>>>>>>>> +      * Then it is possible to support "rpmsg-client-sample*", i.e:
>>>>>>>>>> +      *      rpmsg-client-sample
>>>>>>>>>> +      *      rpmsg-client-sample_instance0
>>>>>>>>>> +      *      rpmsg-client-sample_instance1
>>>>>>>>>> +      *      ...
>>>>>>>>>> +      *      rpmsg-client-sample_instanceX
>>>>>>>>>> +      */
>>>>>>>>>> +     return strncmp(id->name, rpdev->id.name, len) == 0;
>>>>>>>>>>  }
>>>>>>>>>>
>>>>>>>>>>  /* match rpmsg channel and rpmsg driver */
>>>>>>>>>>
>>>>>>>>>
>>>>>>>
>>>
>