Re: [9fans] "gpio device" for Plan 9

[Skip Tavakkolian <skip.tavakkolian@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2367 bytes --]

i won't answer for Erik, but...

there is nothing magic about "long" reads. basically you create multiple
proc's (i.e. rfork) and dedicate one to the reader function. ioproc(2)
library takes care of the housekeeping nicely; also the man page has an
example.



On Tue, Dec 31, 2013 at 11:37 AM, Shane Morris <edgecomberts@gmail.com>wrote:

> Erik,
>
> Just for the purposes of edification (and curiosity), are you able to
> elaborate on "long reads"? Its understandable such a scheme would be
> implemented in the network drivers, but how exactly does it work, as
> opposed to a polling scheme or an ISR? I will, of course, Google in a sec
> as well.
>
> Many thanks!
>
>
> On Wed, Jan 1, 2014 at 6:18 AM, erik quanstrom <quanstro@quanstro.net>wrote:
>
>> On Tue Dec 31 12:47:30 EST 2013, krystian.lew@gmail.com wrote:
>> > Thank you for the feedback, i think "ctl" file and numbering scheme
>> > selection could do the job.  And maybe it could help to establish
>> > reasonable base for SPI and others.
>> >
>> > Is it safe to just generate new dev tree - to return either BCM,
>> > WiringPi or board pin set - based on pin numbering scheme selection
>> > made by user?  What will happen if a process would try o read/write
>> > from/to pin when numbering scheme is changed?  I tried to look at
>> > devproc.c (what would happen when process dies and something is
>> > reading its /proc entries) but i can�♯ see any specific
>> > precautions there.
>>
>> (sorry about the funny formatting.  the header specifies
>>         Content-Type: text/plain; charset=iso-2022-jp-2
>> which might be the same as iso-2022-jp, but i haven't tracked this down
>> yet.)
>>
>> there is a 1 character argument to attach.  you can avoid the issue
>> by letting the attach argument specify which scheme you'd like, e.g.:
>>
>>         mount -a '#Gx' /dev
>>
>> > Regarding ISRs - this is not implemented yet.  Polling at the moment
>> > is the only option.  But maybe "events” file, with data
>> > populated by interrupt routine would be the answer.  Is it correct
>> > Plan 9 way of doing things?  QIO looks very suitable for this purpose.
>>
>> "long" reads are an established way to avoid polling.  plan 9
>> was doing this long before i'd heard the term.  the network drivers
>> work this way.
>>
>> - erik
>>
>>
>

[-- Attachment #2: Type: text/html, Size: 3374 bytes --]