Re: [PATCH 2/2] media: ov9282: enable single-read I2C transactions

[Alexander Stein]

Hi everyone,

Am Dienstag, 21. Juli 2026, 17:32:12 CEST schrieb Laurent Pinchart:
> Hi Richard,
> 
> On Tue, Jul 21, 2026 at 05:15:06PM +0200, Richard Leitner wrote:
> > On Tue, Jul 21, 2026 at 03:38:26PM +0200, Alexander Stein wrote:
> > > Am Dienstag, 21. Juli 2026, 15:25:29 CEST schrieb Richard Leitner:
> > > > Some OV9282 camera modules, such as those from Vision Components, fail
> > > > multi-byte register reads with the default CCI I2C regmap access pattern
> > > > and only work when single-register read transactions are used. This
> > > > results in the driver being unable to probe due to a chip id mismatch.
> > > > 
> > > > As the driver has no reliable way to identify affected modules enable
> > > > use_single_read for OV9282 unconditionally when creating the CCI regmap.
> > > 
> > > What about a property in the DT instead? In DT it should be clear if a
> > > Vision Components camera is used.
> > 
> > I've talked to Laurent about this topic (I guess a year ago or so...)
> > and he pointed me to his downstream work regarding the vc-mipi driver
> > (which I am currently also using):
> > 
> > https://gitlab.com/ideasonboard/nxp/linux/-/commits/v7.0/sensors/vcmipi

Yep, I've also been using that one. And yes, this I2C access peculiarity has
been an issue for a long time already.

> > There is a new driver (and therefore DT compatible) for the controller
> > (without the sensor) on the Vision Components board implemented. This
> > separates basically the "regulator" part of the VC microprocessor from
> > the "forward I2C traffic to the sensor" part. AFAICT...
> > 
> > Talking about that separation I vaguely remember he argued against a
> > new device-tree property or compatible for the vcmipi variants of the
> > ov9282. Therefore I went for the "easy" approach of just enabling single
> > register reads.
> > 
> > Nonetheless if I understood this wrong or "things" changed, please feel
> > free to correct me.
> > 
> > Just to be clear: I'm also perfectly fine with introducing a new
> > "vcmipi,ov9282" compatible or something like a "quirk-vision-components"
> > property if that's preferred.
> 
> My main concern is that I don't want to see this being handled manually
> in drivers. Whatever solution we decide to implement (I proposed two in
> a separate e-mail in this thread) should not require modifying image
> sensor drivers to implement logic specific to the Vision Components
> module. I'm fine relying on drivers using specific helpers (for instance
> requiring drivers to be ported to regmap or v4l2-cci in order to support
> the quirk).

I had the idea of a generic property like 'i2c-disable-autoincrement' usable
on all i2c devices. This property might be parsed in regmap-i2c (or even
i2c-core) so it is enabled and available for all i2c devices without
modifying any driver.

Best regards,
Alexander

> > > > Tested on an i.M8MP system with OV9282 modules from Vision Components.
> > > > 
> > > > Signed-off-by: Richard Leitner <richard.leitner@xxxxxxxxx>
> > > > ---
> > > >  drivers/media/i2c/ov9282.c | 7 ++++++-
> > > >  1 file changed, 6 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/drivers/media/i2c/ov9282.c b/drivers/media/i2c/ov9282.c
> > > > index 5b6f897a74fcd..8267a0a9c6f2d 100644
> > > > --- a/drivers/media/i2c/ov9282.c
> > > > +++ b/drivers/media/i2c/ov9282.c
> > > > @@ -1283,6 +1283,11 @@ static int ov9282_probe(struct i2c_client *client)
> > > >  	struct ov9282 *ov9282;
> > > >  	int ret;
> > > >  
> > > > +	const struct cci_regmap_config config = {
> > > > +		.reg_addr_bits = 16,
> > > > +		.use_single_read = true,
> > > > +	};
> > > > +
> > > >  	ov9282 = devm_kzalloc(&client->dev, sizeof(*ov9282), GFP_KERNEL);
> > > >  	if (!ov9282)
> > > >  		return -ENOMEM;
> > > > @@ -1301,7 +1306,7 @@ static int ov9282_probe(struct i2c_client *client)
> > > >  		return ret;
> > > >  	}
> > > >  
> > > > -	ov9282->regmap = devm_cci_regmap_init_i2c(client, 16);
> > > > +	ov9282->regmap = devm_cci_regmap_init_i2c_cfg(client, &config);
> > > >  	if (IS_ERR(ov9282->regmap))
> > > >  		return dev_err_probe(ov9282->dev, PTR_ERR(ov9282->regmap),
> > > >  				     "Failed to init CCI\n");
> 
> 


-- 
TQ-Systems GmbH | Mühlstraße 2, Gut Delling | 82229 Seefeld, Germany
Amtsgericht München, HRB 105018
Geschäftsführer: Detlef Schneider, Rüdiger Stahl, Stefan Schneider
http://www.tq-group.com/