Re: [PATCH V4 0/4] Add TIOCM Signals support for RPMSG char devices

From: Arnaud Pouliquen
Date: Tue Oct 09 2018 - 12:02:46 EST


hello Bjorn,

On 10/08/2018 06:23 PM, Bjorn Andersson wrote:
> On Mon 08 Oct 06:08 PDT 2018, Arnaud Pouliquen wrote:
>
>> Hi Arun, Bjorn,
>>
>> On 10/08/2018 08:38 AM, Arun Kumar Neelakantam wrote:
>>> Glink transport support signals to exchange state notification between
>>> local and remote side clients. Adding support to send/receive the signal
>>> command and notify the clients through callback and POLL notification.
>>
>> Please correct me if i'm wrong...My concern here is that this patchset
>> implements a rpmsg service in the rpmsg core. I would separate this from
>> the rpmsg core, as this is not part of the rpmsg protocol but seems
>> linked to the serial protocol itself.
>> Could it be implemented in rpmsg_char, using a dedicated channel...?
>>
>
> rpmsg_char does expose a rpmsg channel (be it virtio, smd or glink) to
> user space. This patch series add support for invoking TIOCMGET and
> TIOCMSET on these channels.

I'm not familiar with this concept, that could explain that i don't
understand this patchset...

TIOCMGET and TIOCMSET is related to the serial/console flow control, to
control remote modem/processor, right?

it seems be implemented only to support the rpmsg_char ioctl interface.
When i have a look to the glink code, signal is treated by a message
sent to remote processor. Therefore it seems that it could be treated as
a service on top of rpmsg (so treat it in rpmsg_char instead of extend
the rpmsg protocol to treat it in glink driver).

>
> In addition to adding the client side of this to rpmsg_char it provides
> this support for glink, but the same mechanism exists in smd - while
> this is not supported (today) by the virtio rpmsg.
This is my main concern, i would like to be sure that this service is
not related to specific needs introduced by the rmpsg char implementation:
In this case this should not be part of the rpmsg core but perhaps some
ops directly provided to the rpmsg_char on registration
(rpmsg_chrdev_register_device?)...

>
>
> I'm uncertain of how we could implement this mechanism for virtio rpmsg,
> given that it as a transport doesn't really have a concept of
> channels/flows - but it's really useful to have!
>
> PS. rpmsg_set_signals() can be called from any rpmsg device to perform
> flow control of the communication channel.

For my point of view this patch-set extends the rpmsg protocol to add
channel flow control.
Does it make sense to have a flow control in rpmsg protocol?
if yes, should it be linked to a channel or to the remote processor itself?

Extra comment: associated documentation update is missing.

Regards,
Arnaud

>
> Regards,
> Bjorn
>
>> Regards
>> Arnaud
>>
>>>
>>> Changes since v1:
>>> - Split the patches as per functional areas like core, char, glink
>>> - Add set, clear mask for TIOCMSET
>>> - Merge the char signal callback and POLLPRI patches
>>>
>>> Changes since v2:
>>> - Modify the rpmsg_get_signals function prototype
>>>
>>> Changes since v3:
>>> - Correct the TICOMGET case handling as per new rpmsg_get_signals prototype
>>> - Update the rpmsg_get_signals function header
>>>
>>> Arun Kumar Neelakantam (4):
>>> rpmsg: core: Add signal API support
>>> rpmsg: glink: Add support to handle signals command
>>> rpmsg: char: Add TIOCMGET/TIOCMSET ioctl support
>>> rpmsg: char: Add signal callback and POLLPRI support
>>>
>>> drivers/rpmsg/qcom_glink_native.c | 126 ++++++++++++++++++++++++++++++++++++++
>>> drivers/rpmsg/rpmsg_char.c | 74 +++++++++++++++++++++-
>>> drivers/rpmsg/rpmsg_core.c | 41 +++++++++++++
>>> drivers/rpmsg/rpmsg_internal.h | 5 ++
>>> include/linux/rpmsg.h | 26 ++++++++
>>> 5 files changed, 269 insertions(+), 3 deletions(-)
>>>