Message-ID: <525E37BF.4040303@onera.fr>
Date:	Wed, 16 Oct 2013 08:52:47 +0200
From:	Paul Chavent <Paul.Chavent@...ra.fr>
To:	Andrew Morton <akpm@...ux-foundation.org>,
	Rodolfo Giometti <giometti@...eenne.com>
CC:	linux-kernel@...r.kernel.org,
	Alexander Gordeev <agordeev@...hat.com>
Subject: Re: [PATCH] pps : add non blocking option to PPS_FETCH ioctl.



On 10/15/2013 09:55 PM, Andrew Morton wrote:
> On Tue, 15 Oct 2013 12:43:50 +0200 Rodolfo Giometti <giometti@...eenne.com> wrote:
>
>> On Fri, Oct 11, 2013 at 12:47:20PM -0700, Andrew Morton wrote:
>>> On Fri, 11 Oct 2013 14:40:32 +0200 Paul Chavent <Paul.Chavent@...ra.fr> wrote:
>>>
>>>> The PPS_FETCH ioctl is blocking still the reception of a PPS
>>>> event. But, in some case, one may immediately need the last event
>>>> date. This patch allow to get the result of PPS_FETCH if the device
>>>> has the O_NONBLOCK flag set.
>>>
>>> Are the PPS ioctls actually documented anywhere?
>>> Documentation/pps/pps.txt is silent.
>>>
>>> That's a shame, because it would be nice to have a formal description
>>> of the the PPS_FETCH semantics which leads to an understanding of why
>>> things are the way they are, how PPS_FETCH is supposed to be used, etc.
>>>
>>> Also, the presence of such documentation would permit me to bug you for
>>> not having updated it!  We need *some* channel for telling people about
>>> the driver, and updates to it.  Maybe linuxpps.org has it somewhere,
>>> but I couldn't immediately find it.
>>
>> Hi Andrew! Actually RFC 2783 doesn't use ioctls to get access to PPS
>> data but it defines some functions. LinuxPPS, that is the Linux PPS
>> implementation, uses ioctls to implement these functions.
>>
>> If you like having an idea about how these functions are implemented
>> into LinuxPPS, you can see here:
>>
>>     http://www.linuxpps.org/gitweb/?p=pps-tools;a=blob;f=timepps.h;h=d2628d2d061ea2a3623e57990d9ada62623773cf;hb=5980a044bcdb4c1d1a8b1ecff986fa63719519b3
>>
>>> Your implementation requires that the file be opened non-blocking.  But
>>> I'd have thought that adding a new and separate ioctl mode would be a
>>> cleaner and more flexible implementation - that way an app which wants
>>> both blocking and non-blocking behaviour doesn't need to open the file
>>> twice.
>>>
>>> Also, this is actually a non-backward-compatible change for any
>>> application which happened to be opening the file with O_NONBLOCK!
>>> Hopefully there aren't any such applications...
>>
>> The major application that I know using these layer is NTPD... however
>> all RFC compliant applications should not use ioctls to get access the
>> PPS data and this patch should be a "special" case.
>>
>
> Thanks, but this doesn't really address my concerns.
>
> - Where, if anywhere, is the Linux PPS API documented and how do we
>    communicate changes such as this one to kernel users?
>
> - Why require open(O_NONBLOCK) for this instead of adding a separate
>    ioctl?
>

I would also prefer the separate ioctl. As you said it, it's a bit 
annoying to switch from blocking mode to non blocking mode if we need 
both mode. But i was not sure about the preferences of the maintainer : 
(i) change the api, or (ii) change the behavior with a widely supported 
interface (O_NONBLOCK).

I'm certainly not the best person to make the final decision, but i 
would like to help you if you need me (write doc, or change this patch).

Cheers.

Paul.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists