Re: [PATCH v2 4/5] media: i2c: ov5645: Return zero for s_stream(0)

[Sakari Ailus <sakari.ailus@linux.intel.com>]

On Mon, Oct 17, 2022 at 12:03:17AM +0300, Laurent Pinchart wrote:
> Hi Sakari,
> 
> On Sun, Oct 16, 2022 at 08:19:40PM +0000, Sakari Ailus wrote:
> > On Sun, Oct 16, 2022 at 02:23:13AM +0300, Laurent Pinchart wrote:
> > > On Sat, Oct 15, 2022 at 09:35:12PM +0000, Sakari Ailus wrote:
> > > > On Sat, Oct 15, 2022 at 09:25:37AM +0300, Laurent Pinchart wrote:
> > > > > On Fri, Oct 14, 2022 at 07:34:58PM +0100, Prabhakar wrote:
> > > > > > From: Lad Prabhakar <prabhakar.mahadev-lad.rj@bp.renesas.com>
> > > > > > 
> > > > > > Always return zero while stopping the stream as the caller will ignore the
> > > > > > return value.
> > > > > > 
> > > > > > This patch drops checking the return value of ov5645_write_reg() and
> > > > > > continues further in the code path while stopping stream. The user anyway
> > > > > > gets an error message in case ov5645_write_reg() fails.
> > > > > 
> > > > > Continuing all the way to pm_runtime_put() is fine, but I don't think
> > > > > the function should return 0. It's not up to the driver to decide if a
> > > > > failure would be useful to signal to the caller or not.
> > > > 
> > > > If the function returns an error when disabling streaming, what is the
> > > > expected power state of the device after this?
> > > 
> > > That's up to us to decide :-)
> > > 
> > > > The contract between the caller and the callee is that the state is not
> > > > changed if there is an error.
> > > 
> > > For most APIs, but that's not universal.
> > > 
> > > > This is a special case as very few callers
> > > > check the return value for streamoff operation and those that do generally
> > > > just print something. I've never seen a caller trying to prevent streaming
> > > > off in this case, for instance.
> > > 
> > > I think the stream off call should proceed and try to power off the
> > > device even if an error occurs along the way, i.e. it shouldn't return
> > > upon the first detected error.
> > > 
> > > > Of course we could document that streaming off always counts as succeeded
> > > > (e.g. decreasing device's runtime PM usage_count) while it could return an
> > > > informational error code. But I wonder if anyone would ever benefit from
> > > > that somehow. :-)
> > > 
> > > I think it could be useful to propagate errors up to inform the user
> > > that something wrong happened. That would involve fixing lots of drivers
> > > along the call chain though, so there's no urgency for the ov5645 to do
> > > so, but isn't it better to propagate the error code instead of hiding
> > > the issue ?
> > 
> > I also don't think hiding the issue would be the best thing to do, but that
> > wouldn't likely be a big problem either.
> > 
> > How about printing a warning in the wrapper while returning zero to the
> > original caller? This would keep the API intact while still leaving a trace
> > on something failing. Of course the driver is also free to print whatever
> > messages it likes.
> 
> While I think error propagation could be more useful in the long run,
> printing a message in the wrapper is a good idea. I like centralized
> error handling, it has a tendency to go wrong when left to individual
> drivers.

I can send a patch...

-- 
Sakari Ailus