Re: [PATCH] media: staging/ipu7: avoid division by 64-bit value

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

Hi David,

On Fri, Jul 11, 2025 at 09:23:29AM +0100, David Laight wrote:
> On Wed,  9 Jul 2025 16:56:56 +0200
> Arnd Bergmann <arnd@kernel.org> wrote:
> 
> > From: Arnd Bergmann <arnd@arndb.de>
> > 
> > On 32-bit targets, this causes a link failure:
> > 
> > x86_64-linux-ld: drivers/staging/media/ipu7/ipu7-isys-csi-phy.o: in function `ipu7_isys_phy_config':
> > ipu7-isys-csi-phy.c:(.text+0x1509): undefined reference to `__udivdi3'
> > 
> > Note that this does not divide a 64-bit number by a 32-bit one as usual,
> > but the other way round, which is something that the compiler should
> > really be able to figure out but does not (as of gcc-15).
> > 
> > A few lines higher, a similar division is done using the incorrect div_u64()
> > that truncates the 64-bit divisor to 32 bits.
> > 
> > Change both to use the safe but slow div64_u64() helper.
> > 
> > Fixes: a516d36bdc3d ("media: staging/ipu7: add IPU7 input system device driver")
> > Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> > ---
> >  drivers/staging/media/ipu7/ipu7-isys-csi-phy.c | 8 ++++----
> >  1 file changed, 4 insertions(+), 4 deletions(-)
> > 
> > diff --git a/drivers/staging/media/ipu7/ipu7-isys-csi-phy.c b/drivers/staging/media/ipu7/ipu7-isys-csi-phy.c
> > index 4407750c7f40..b8c5db7ae300 100644
> > --- a/drivers/staging/media/ipu7/ipu7-isys-csi-phy.c
> > +++ b/drivers/staging/media/ipu7/ipu7-isys-csi-phy.c
> > @@ -734,6 +734,7 @@ static void ipu7_isys_cphy_config(struct ipu7_isys *isys, u8 id, u8 lanes,
> >  	u16 reg;
> >  	u16 val;
> >  	u32 i;
> > +	u64 r64;
> >  	u32 r;
> >  
> >  	if (is_ipu7p5(isys->adev->isp->hw_ver))
> > @@ -806,8 +807,8 @@ static void ipu7_isys_cphy_config(struct ipu7_isys *isys, u8 id, u8 lanes,
> >  		dwc_phy_write_mask(isys, id, reg, 2, 0, 2);
> >  	}
> >  
> > -	deass_thresh = (u16)div_u64_rem(7 * 1000 * 6, mbps * 5U, &r) + 1;
> > -	if (r != 0)
> > +	deass_thresh = (u16)div64_u64_rem(7 * 1000 * 6, mbps * 5U, &r64) + 1;
> > +	if (r64 != 0)
> >  		deass_thresh++;
> 
> While 'mbps' is presumably u64, it can't be big.
> So this can just be:
> 	deass_threas = roundup(7 * 1000 * 6 / 5, (u32)mbps) + 1;
> 
> 
> >  
> >  	reg = CORE_DIG_RW_TRIO0_2;
> > @@ -815,8 +816,7 @@ static void ipu7_isys_cphy_config(struct ipu7_isys *isys, u8 id, u8 lanes,
> >  		dwc_phy_write_mask(isys, id, reg + 0x400 * i,
> >  				   deass_thresh, 0, 7);
> >  
> > -	delay_thresh =
> > -		((224U - (9U * 7U)) * 1000U) / (5U * mbps) - 7U;
> > +	delay_thresh = div64_u64((224U - (9U * 7U)) * 1000U, 5U * mbps) - 7u;
> 
> That one just needs a (u32) cast, although the 5 can be moved.

I agree in principle but this should be done separately from fixing the bug
which is what this patch does. mbps is u64 elsewhere, too, and there
doesn't seem to be a need for that.

-- 
Regards,

Sakari Ailus