Re: [RFC PATCH v1] iio: trigger: move trig->owner init to trigger allocate() stage

From: Jonathan Cameron
Date: Mon Jun 06 2022 - 11:13:18 EST


On Mon, 6 Jun 2022 11:37:42 +0000
Dmitry Rokosov <DDRokosov@xxxxxxxxxxxxxx> wrote:

> Hello Jonathan,
>
> Thank you for comments. I have a several questions about the flow,
> please find them below.
>
> On Sat, Jun 04, 2022 at 02:59:55PM +0100, Jonathan Cameron wrote:
> > On Wed, 1 Jun 2022 17:48:32 +0000
> > Dmitry Rokosov <DDRokosov@xxxxxxxxxxxxxx> wrote:
> >
> > > To provide a new IIO trigger to the IIO core, usually driver executes the
> > > following pipeline: allocate()/register()/get(). Before, IIO core assigned
> > > trig->owner as a pointer to the module which registered this trigger at
> > > the register() stage. But actually the trigger object is owned by the
> > > module earlier, on the allocate() stage, when trigger object is
> > > successfully allocated for the driver.
> > >
> > > This patch moves trig->owner initialization from register()
> > > stage of trigger initialization pipeline to allocate() stage to
> > > eliminate all misunderstandings and time gaps between trigger object
> > > creation and owner acquiring.
> > >
> > > Signed-off-by: Dmitry Rokosov <ddrokosov@xxxxxxxxxxxxxx>
> >
> > Hi Dmitry,
> >
> > I 'think' this is fine, but its in the high risk category that I'd like
> > to keep it on list for a few weeks before applying.
> >
>
> Could you please explain what it means? Do you have some testing branch
> with such dangerous patches or do we need just to wait other developers
> for more points of view? Thanks in advance.

The second - so far I haven't applied it anywhere.

>
> > Note I'm still keen that in general we keep the flow such that
> > we do allocate()/register()/get() as there is no guarantee that the get()
> > will never do anything that requires the trigger to be registered, even
> > though that is true today. Which is another way of saying I'm still
> > keen we fix up any cases that sneak in after your fix up set dealt with
> > the current ones.
>
> I fully agree with you. I suppose to resolve such a problem we need to
> have some indicators that the trigger is already registered or not.
> From my point of view, trig->list entry fits well to answer this question.
> Trigger is added to the global IIO triggers list during register()
> execution, so we can just check that entry is not empty to make sure that
> trigger is registered.
>
> I've sent a v2 patch version, where I use trig->list entry empty status to
> warn it:
>
> https://lore.kernel.org/linux-iio/20220606111316.19265-1-ddrokosov@xxxxxxxxxxxxxx/

Great!

Jonathan

>
> >
> > Thanks for following up on this!
> >
> > Jonathan
> >
>