Hi Daniel,
On Fri, May 18, 2018 at 10:32:43AM +0200, Daniel Mack wrote:
> On Thursday, May 17, 2018 10:53 AM, Maxime Ripard wrote:
> > From: Samuel Bobrowicz <sam@xxxxxxxxxxxxxxxxxx>
> >
> > The current code, when changing the mode and changing the scaling or
> > sampling parameters, will look at the horizontal and vertical total size,
> > which, since 5999f381e023 ("media: ov5640: Add horizontal and vertical
> > totals") has been moved from the static register initialization to after
> > the mode change.
> >
> > That means that the values are no longer set up before the code retrieves
> > them, which is obviously a bug.
>
> I tried 'for-4.18-5' before your patch set now and noticed it didn't work. I
> then bisected the regression down to the same commit that you mentioned
> above.
>
> The old code (before 5999f381e023) used to have the timing registers
> embedded in a large register blob. Omitting them during the bulk upload and
> writing them later doesn't work here, even if the values are the same. It
> seems that the order in which registers are written matters.
>
> Hence your patch in this email doesn't fix it for me either. I have to move
> ov5640_set_timings() before ov5640_load_regs() to revive my platform.
That's kind of weird, it definitely works for parallel. Do you have a
bit more details on what doesn't work? What commands / apps are you
running?
Maxime
--
Maxime Ripard, Bootlin (formerly Free Electrons)
Embedded Linux and Kernel engineering
https://bootlin.com
Attachment:
signature.asc
Description: PGP signature
