[PATCHv3 1/1 RESEND] media: imx: Round line size to 4 bytes

[PATCHv3 1/1 RESEND] media: imx: Round line size to 4 bytes

[Dorota Czaplejewicz]

[-- Attachment #1: Type: text/plain, Size: 1420 bytes --]

Section 13.7.6.13 "CSI Image Parameter Register" of the
i.MX 8M Quad Applications Processors Reference Manual
states that the line size should be divisible by 8 bytes.
However, the hardware also accepts sizes divisible by 4 bytes,
which are needed to fully utilize the S5K3L6XX sensors.

This patch accepts line sizes divisible by 4 bytes.

Signed-off-by: Dorota Czaplejewicz <dorota.czaplejewicz@puri.sm>
---
Hello,

the Librem 5 is using an out-of-tree driver for s5k3l6xx, and rounding to 4 is optimal to operate it.

This revision improves the commit message.

Cheers,
Dorota Czaplejewicz

 drivers/staging/media/imx/imx7-media-csi.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/staging/media/imx/imx7-media-csi.c b/drivers/staging/media/imx/imx7-media-csi.c
index a0553c24cce4..af821b410c3f 100644
--- a/drivers/staging/media/imx/imx7-media-csi.c
+++ b/drivers/staging/media/imx/imx7-media-csi.c
@@ -999,10 +999,10 @@ static int imx7_csi_mbus_fmt_to_pix_fmt(struct v4l2_pix_format *pix,
 	}
 
 	/* Round up width for minimum burst size */
-	width = round_up(mbus->width, 8);
+	width = round_up(mbus->width, 4);
 
 	/* Round up stride for IDMAC line start address alignment */
-	stride = round_up((width * cc->bpp) >> 3, 8);
+	stride = round_up((width * cc->bpp) >> 3, 4);
 
 	pix->width = width;
 	pix->height = mbus->height;
-- 
2.37.3


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCHv3 1/1 RESEND] media: imx: Round line size to 4 bytes

[Laurent Pinchart]

Hi Dorota,

Thank you for the patch.

On Sun, Oct 16, 2022 at 04:39:28PM +0200, Dorota Czaplejewicz wrote:
> Section 13.7.6.13 "CSI Image Parameter Register" of the
> i.MX 8M Quad Applications Processors Reference Manual
> states that the line size should be divisible by 8 bytes.
> However, the hardware also accepts sizes divisible by 4 bytes,
> which are needed to fully utilize the S5K3L6XX sensors.
> 
> This patch accepts line sizes divisible by 4 bytes.

The patch itself looks fine, but I always worry when we depart from the
documentation (sometimes that's a worry about documentation inaccuracies
:-)). In this particular case, we need to test the change on i.MX7 too,
as the alignment constraints could be platform-specific.

I would also very much appreciate feedback from NXP on this. There's a
risk of ill side-effects that I would prefer ruling out very clearly.

> Signed-off-by: Dorota Czaplejewicz <dorota.czaplejewicz@puri.sm>
> ---
> Hello,
> 
> the Librem 5 is using an out-of-tree driver for s5k3l6xx, and rounding to 4 is optimal to operate it.
> 
> This revision improves the commit message.
> 
> Cheers,
> Dorota Czaplejewicz
> 
>  drivers/staging/media/imx/imx7-media-csi.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/staging/media/imx/imx7-media-csi.c b/drivers/staging/media/imx/imx7-media-csi.c
> index a0553c24cce4..af821b410c3f 100644
> --- a/drivers/staging/media/imx/imx7-media-csi.c
> +++ b/drivers/staging/media/imx/imx7-media-csi.c
> @@ -999,10 +999,10 @@ static int imx7_csi_mbus_fmt_to_pix_fmt(struct v4l2_pix_format *pix,
>  	}
>  
>  	/* Round up width for minimum burst size */
> -	width = round_up(mbus->width, 8);
> +	width = round_up(mbus->width, 4);
>  
>  	/* Round up stride for IDMAC line start address alignment */
> -	stride = round_up((width * cc->bpp) >> 3, 8);
> +	stride = round_up((width * cc->bpp) >> 3, 4);
>  
>  	pix->width = width;
>  	pix->height = mbus->height;

-- 
Regards,

Laurent Pinchart

Re: [PATCHv3 1/1 RESEND] media: imx: Round line size to 4 bytes

[Dorota Czaplejewicz]

[-- Attachment #1: Type: text/plain, Size: 2523 bytes --]

Hi,

On Mon, 17 Oct 2022 11:28:57 +0300
Laurent Pinchart <laurent.pinchart@ideasonboard.com> wrote:

> Hi Dorota,
> 
> Thank you for the patch.
> 
> On Sun, Oct 16, 2022 at 04:39:28PM +0200, Dorota Czaplejewicz wrote:
> > Section 13.7.6.13 "CSI Image Parameter Register" of the
> > i.MX 8M Quad Applications Processors Reference Manual
> > states that the line size should be divisible by 8 bytes.
> > However, the hardware also accepts sizes divisible by 4 bytes,
> > which are needed to fully utilize the S5K3L6XX sensors.
> > 
> > This patch accepts line sizes divisible by 4 bytes.  
> 
> The patch itself looks fine, but I always worry when we depart from the
> documentation (sometimes that's a worry about documentation inaccuracies
> :-)). In this particular case, we need to test the change on i.MX7 too,
> as the alignment constraints could be platform-specific.
> 
I don't have access to any i.MX7 hardware, unfortunately. I can help anyone who does and wants to test this.

> I would also very much appreciate feedback from NXP on this. There's a
> risk of ill side-effects that I would prefer ruling out very clearly.

I posted a question on the NXP forum: https://community.nxp.com/t5/i-MX-Processors/i-MX8MQ-CSI-line-size-constraint-documentation-mistake/m-p/1538629#M196448

--Dorota
> 
> > Signed-off-by: Dorota Czaplejewicz <dorota.czaplejewicz@puri.sm>
> > ---
> > Hello,
> > 
> > the Librem 5 is using an out-of-tree driver for s5k3l6xx, and rounding to 4 is optimal to operate it.
> > 
> > This revision improves the commit message.
> > 
> > Cheers,
> > Dorota Czaplejewicz
> > 
> >  drivers/staging/media/imx/imx7-media-csi.c | 4 ++--
> >  1 file changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/staging/media/imx/imx7-media-csi.c b/drivers/staging/media/imx/imx7-media-csi.c
> > index a0553c24cce4..af821b410c3f 100644
> > --- a/drivers/staging/media/imx/imx7-media-csi.c
> > +++ b/drivers/staging/media/imx/imx7-media-csi.c
> > @@ -999,10 +999,10 @@ static int imx7_csi_mbus_fmt_to_pix_fmt(struct v4l2_pix_format *pix,
> >  	}
> >  
> >  	/* Round up width for minimum burst size */
> > -	width = round_up(mbus->width, 8);
> > +	width = round_up(mbus->width, 4);
> >  
> >  	/* Round up stride for IDMAC line start address alignment */
> > -	stride = round_up((width * cc->bpp) >> 3, 8);
> > +	stride = round_up((width * cc->bpp) >> 3, 4);
> >  
> >  	pix->width = width;
> >  	pix->height = mbus->height;  
> 


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCHv3 1/1 RESEND] media: imx: Round line size to 4 bytes

[Dorota Czaplejewicz]

[-- Attachment #1: Type: text/plain, Size: 2330 bytes --]

Hi all,

On Mon, 17 Oct 2022 16:30:59 +0200
Dorota Czaplejewicz <dorota.czaplejewicz@puri.sm> wrote:

> Hi,
> 
> On Mon, 17 Oct 2022 11:28:57 +0300
> Laurent Pinchart <laurent.pinchart@ideasonboard.com> wrote:
> 

[snip]

> > I would also very much appreciate feedback from NXP on this. There's a
> > risk of ill side-effects that I would prefer ruling out very clearly.  
> 
> I posted a question on the NXP forum: https://community.nxp.com/t5/i-MX-Processors/i-MX8MQ-CSI-line-size-constraint-documentation-mistake/m-p/1538629#M196448
> 

This question received a reply from NXP today: https://community.nxp.com/t5/i-MX-Processors/i-MX8MQ-CSI-line-size-constraint-documentation-mistake/m-p/1546872/highlight/true#M197067

Quoting NXP's answer:

> Do I understand correctly, that streams divisible by 4×4 will work on other i.MX8 hardware too? - Yes
> Will those kind of resolutions work on i.MX7 series CSI hardware? - Yes

which implies no bad side effects the way I read it. Is this sufficient for the kernel?

Cheers,
Dorota

> >   
> > > Signed-off-by: Dorota Czaplejewicz <dorota.czaplejewicz@puri.sm>
> > > ---
> > > Hello,
> > > 
> > > the Librem 5 is using an out-of-tree driver for s5k3l6xx, and rounding to 4 is optimal to operate it.
> > > 
> > > This revision improves the commit message.
> > > 
> > > Cheers,
> > > Dorota Czaplejewicz
> > > 
> > >  drivers/staging/media/imx/imx7-media-csi.c | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/staging/media/imx/imx7-media-csi.c b/drivers/staging/media/imx/imx7-media-csi.c
> > > index a0553c24cce4..af821b410c3f 100644
> > > --- a/drivers/staging/media/imx/imx7-media-csi.c
> > > +++ b/drivers/staging/media/imx/imx7-media-csi.c
> > > @@ -999,10 +999,10 @@ static int imx7_csi_mbus_fmt_to_pix_fmt(struct v4l2_pix_format *pix,
> > >  	}
> > >  
> > >  	/* Round up width for minimum burst size */
> > > -	width = round_up(mbus->width, 8);
> > > +	width = round_up(mbus->width, 4);
> > >  
> > >  	/* Round up stride for IDMAC line start address alignment */
> > > -	stride = round_up((width * cc->bpp) >> 3, 8);
> > > +	stride = round_up((width * cc->bpp) >> 3, 4);
> > >  
> > >  	pix->width = width;
> > >  	pix->height = mbus->height;    
> >   
> 


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCHv3 1/1 RESEND] media: imx: Round line size to 4 bytes

[Laurent Pinchart]

Hi Dorota,

On Tue, Nov 01, 2022 at 01:38:34PM +0100, Dorota Czaplejewicz wrote:
> On Mon, 17 Oct 2022 16:30:59 +0200 Dorota Czaplejewicz wrote:
> > On Mon, 17 Oct 2022 11:28:57 +0300 Laurent Pinchart wrote:
> 
> [snip]
> 
> > > I would also very much appreciate feedback from NXP on this. There's a
> > > risk of ill side-effects that I would prefer ruling out very clearly.  
> > 
> > I posted a question on the NXP forum: https://community.nxp.com/t5/i-MX-Processors/i-MX8MQ-CSI-line-size-constraint-documentation-mistake/m-p/1538629#M196448
> > 
> 
> This question received a reply from NXP today: https://community.nxp.com/t5/i-MX-Processors/i-MX8MQ-CSI-line-size-constraint-documentation-mistake/m-p/1546872/highlight/true#M197067
> 
> Quoting NXP's answer:
> 
> > Do I understand correctly, that streams divisible by 4×4 will work on other i.MX8 hardware too? - Yes
> > Will those kind of resolutions work on i.MX7 series CSI hardware? - Yes
> 
> which implies no bad side effects the way I read it. Is this
> sufficient for the kernel?

Ideally I'd like to test that on i.MX7 but I don't think I'll have time
to do so in the near future. I don't want to block the patch on this, so
I'm fine merging it, but I'd like a comment in the code that explains
why it's safe to depart from the documentation.

Also, the discussion on the NXP forum said that a width that is a
multiple of 4 bytes but not a multiple of 8 bytes works only if the
height is also a multiple of 4. I don't see that constraint being
enforced in the code, am I missing something ?

You mentioned in the forum that you tried 1052x779 and that failed. How
did it fail ? Have you also tried 1052x778 ?

> > > > Signed-off-by: Dorota Czaplejewicz <dorota.czaplejewicz@puri.sm>
> > > > ---
> > > > Hello,
> > > > 
> > > > the Librem 5 is using an out-of-tree driver for s5k3l6xx, and
> > > > rounding to 4 is optimal to operate it.
> > > > 
> > > > This revision improves the commit message.
> > > > 
> > > > Cheers,
> > > > Dorota Czaplejewicz
> > > > 
> > > >  drivers/staging/media/imx/imx7-media-csi.c | 4 ++--
> > > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > > 
> > > > diff --git a/drivers/staging/media/imx/imx7-media-csi.c b/drivers/staging/media/imx/imx7-media-csi.c
> > > > index a0553c24cce4..af821b410c3f 100644
> > > > --- a/drivers/staging/media/imx/imx7-media-csi.c
> > > > +++ b/drivers/staging/media/imx/imx7-media-csi.c
> > > > @@ -999,10 +999,10 @@ static int imx7_csi_mbus_fmt_to_pix_fmt(struct v4l2_pix_format *pix,
> > > >  	}
> > > >  
> > > >  	/* Round up width for minimum burst size */
> > > > -	width = round_up(mbus->width, 8);
> > > > +	width = round_up(mbus->width, 4);
> > > >  
> > > >  	/* Round up stride for IDMAC line start address alignment */
> > > > -	stride = round_up((width * cc->bpp) >> 3, 8);
> > > > +	stride = round_up((width * cc->bpp) >> 3, 4);
> > > >  
> > > >  	pix->width = width;
> > > >  	pix->height = mbus->height;    

-- 
Regards,

Laurent Pinchart

Re: [PATCHv3 1/1 RESEND] media: imx: Round line size to 4 bytes

[Dorota Czaplejewicz]

[-- Attachment #1: Type: text/plain, Size: 3701 bytes --]

Hello,

On Thu, 24 Nov 2022 13:30:15 +0200
Laurent Pinchart <laurent.pinchart@ideasonboard.com> wrote:

> Hi Dorota,
> 
> On Tue, Nov 01, 2022 at 01:38:34PM +0100, Dorota Czaplejewicz wrote:
> > On Mon, 17 Oct 2022 16:30:59 +0200 Dorota Czaplejewicz wrote:  
> > > On Mon, 17 Oct 2022 11:28:57 +0300 Laurent Pinchart wrote:  
> > 
> > [snip]
> >   
> > > > I would also very much appreciate feedback from NXP on this. There's a
> > > > risk of ill side-effects that I would prefer ruling out very clearly.    
> > > 
> > > I posted a question on the NXP forum: https://community.nxp.com/t5/i-MX-Processors/i-MX8MQ-CSI-line-size-constraint-documentation-mistake/m-p/1538629#M196448
> > >   
> > 
> > This question received a reply from NXP today: https://community.nxp.com/t5/i-MX-Processors/i-MX8MQ-CSI-line-size-constraint-documentation-mistake/m-p/1546872/highlight/true#M197067
> > 
> > Quoting NXP's answer:
> >   
> > > Do I understand correctly, that streams divisible by 4×4 will work on other i.MX8 hardware too? - Yes
> > > Will those kind of resolutions work on i.MX7 series CSI hardware? - Yes  
> > 
> > which implies no bad side effects the way I read it. Is this
> > sufficient for the kernel?  
> 
> Ideally I'd like to test that on i.MX7 but I don't think I'll have time
> to do so in the near future. I don't want to block the patch on this, so
> I'm fine merging it, but I'd like a comment in the code that explains
> why it's safe to depart from the documentation.
> 
> Also, the discussion on the NXP forum said that a width that is a
> multiple of 4 bytes but not a multiple of 8 bytes works only if the
> height is also a multiple of 4. I don't see that constraint being
> enforced in the code, am I missing something ?

Thanks for pointing it out, I'll have to double-check it.
> 
> You mentioned in the forum that you tried 1052x779 and that failed. How
> did it fail ? Have you also tried 1052x778 ?

I tried with many resolutions when I was checking out the sensor, and came to the conclusion that the line count must be even. Otherwise the last 4 bytes of the buffer would end up corrupted.

My guess is that the copy engine can indeed only do bursts of 8 bytes, but that the whole buffer is copied at once rather than line-by-line.
> 
> > > > > Signed-off-by: Dorota Czaplejewicz <dorota.czaplejewicz@puri.sm>
> > > > > ---
> > > > > Hello,
> > > > > 
> > > > > the Librem 5 is using an out-of-tree driver for s5k3l6xx, and
> > > > > rounding to 4 is optimal to operate it.
> > > > > 
> > > > > This revision improves the commit message.
> > > > > 
> > > > > Cheers,
> > > > > Dorota Czaplejewicz
> > > > > 
> > > > >  drivers/staging/media/imx/imx7-media-csi.c | 4 ++--
> > > > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > > > 
> > > > > diff --git a/drivers/staging/media/imx/imx7-media-csi.c b/drivers/staging/media/imx/imx7-media-csi.c
> > > > > index a0553c24cce4..af821b410c3f 100644
> > > > > --- a/drivers/staging/media/imx/imx7-media-csi.c
> > > > > +++ b/drivers/staging/media/imx/imx7-media-csi.c
> > > > > @@ -999,10 +999,10 @@ static int imx7_csi_mbus_fmt_to_pix_fmt(struct v4l2_pix_format *pix,
> > > > >  	}
> > > > >  
> > > > >  	/* Round up width for minimum burst size */
> > > > > -	width = round_up(mbus->width, 8);
> > > > > +	width = round_up(mbus->width, 4);
> > > > >  
> > > > >  	/* Round up stride for IDMAC line start address alignment */
> > > > > -	stride = round_up((width * cc->bpp) >> 3, 8);
> > > > > +	stride = round_up((width * cc->bpp) >> 3, 4);
> > > > >  
> > > > >  	pix->width = width;
> > > > >  	pix->height = mbus->height;      
> 


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]