drivers:iio:dac make expression evaluation 64-bit

[PATCH] drivers:iio:dac make expression evaluation 64-bit

Posted by Muhammad Usama Anjum 4 years, 6 months ago

Two 32-bit values are being evaluated using 32-bit arithmetic and then
passed to s64 type. It is wrong. Expression should be evaluated using
64-bit arithmetic and then passed.

Fixes: 8f2b54824b ("drivers:iio:dac: Add AD3552R driver support")
Signed-off-by: Muhammad Usama Anjum <usama.anjum@collabora.com>
---
 drivers/iio/dac/ad3552r.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/iio/dac/ad3552r.c b/drivers/iio/dac/ad3552r.c
index 97f13c0b9631..b03d3c7cd4c4 100644
--- a/drivers/iio/dac/ad3552r.c
+++ b/drivers/iio/dac/ad3552r.c
@@ -770,7 +770,7 @@ static void ad3552r_calc_gain_and_offset(struct ad3552r_desc *dac, s32 ch)
 	dac->ch_data[ch].scale_dec = DIV_ROUND_CLOSEST((s64)rem * 1000000,
 							65536);
 
-	dac->ch_data[ch].offset_int = div_s64_rem(v_min * 65536, span, &rem);
+	dac->ch_data[ch].offset_int = div_s64_rem(v_min * 65536L, span, &rem);
 	tmp = (s64)rem * 1000000;
 	dac->ch_data[ch].offset_dec = div_s64(tmp, span);
 }
-- 
2.30.2

Re: [PATCH] drivers:iio:dac make expression evaluation 64-bit

Posted by Andy Shevchenko 4 years, 6 months ago

On Wed, Dec 22, 2021 at 5:59 PM Muhammad Usama Anjum
<usama.anjum@collabora.com> wrote:
>
> Two 32-bit values are being evaluated using 32-bit arithmetic and then
> passed to s64 type. It is wrong. Expression should be evaluated using
> 64-bit arithmetic and then passed.

...

>         dac->ch_data[ch].scale_dec = DIV_ROUND_CLOSEST((s64)rem * 1000000,
>                                                         65536);

Shouldn't the above be fixed as well? Has anybody tried to compile on
32-bit arch this?

> -       dac->ch_data[ch].offset_int = div_s64_rem(v_min * 65536, span, &rem);
> +       dac->ch_data[ch].offset_int = div_s64_rem(v_min * 65536L, span, &rem);
>         tmp = (s64)rem * 1000000;
>         dac->ch_data[ch].offset_dec = div_s64(tmp, span);
>  }

-- 
With Best Regards,
Andy Shevchenko

Re: [PATCH] drivers:iio:dac make expression evaluation 64-bit

Posted by Muhammad Usama Anjum 4 years, 6 months ago

On 12/22/21 11:49 PM, Andy Shevchenko wrote:
> On Wed, Dec 22, 2021 at 5:59 PM Muhammad Usama Anjum
> <usama.anjum@collabora.com> wrote:
>>
>> Two 32-bit values are being evaluated using 32-bit arithmetic and then
>> passed to s64 type. It is wrong. Expression should be evaluated using
>> 64-bit arithmetic and then passed.
> 
> ...
> 
>>         dac->ch_data[ch].scale_dec = DIV_ROUND_CLOSEST((s64)rem * 1000000,
>>                                                         65536);
> 
> Shouldn't the above be fixed as well? Has anybody tried to compile on
> 32-bit arch this?
No, it correct already. In this case, rem is being typecasted to s64 and
then multiplied with a 32-bit number, 1000000. Thus 64-bit arithmetic is
being performed here.

Re: [PATCH] drivers:iio:dac make expression evaluation 64-bit

Posted by Lars-Peter Clausen 4 years, 6 months ago

On 12/23/21 5:34 PM, Muhammad Usama Anjum wrote:
> On 12/22/21 11:49 PM, Andy Shevchenko wrote:
>> On Wed, Dec 22, 2021 at 5:59 PM Muhammad Usama Anjum
>> <usama.anjum@collabora.com> wrote:
>>> Two 32-bit values are being evaluated using 32-bit arithmetic and then
>>> passed to s64 type. It is wrong. Expression should be evaluated using
>>> 64-bit arithmetic and then passed.
>> ...
>>
>>>          dac->ch_data[ch].scale_dec = DIV_ROUND_CLOSEST((s64)rem * 1000000,
>>>                                                          65536);
>> Shouldn't the above be fixed as well? Has anybody tried to compile on
>> 32-bit arch this?
> No, it correct already. In this case, rem is being typecasted to s64 and
> then multiplied with a 32-bit number, 1000000. Thus 64-bit arithmetic is
> being performed here.

What Andy means is that this needs to be DIV_S64_ROUND_CLOSEST() to work 
on 32-bit platforms. But it is clearly unrelated to your change and 
should be in its own patch.

RE: [PATCH] drivers:iio:dac make expression evaluation 64-bit

Posted by Chindris, Mihail 4 years, 5 months ago

> >>> Two 32-bit values are being evaluated using 32-bit arithmetic and
> >>> then passed to s64 type. It is wrong. Expression should be evaluated
> >>> using 64-bit arithmetic and then passed.
> >> ...
> >>
> >>>          dac->ch_data[ch].scale_dec = DIV_ROUND_CLOSEST((s64)rem *
> 1000000,
> >>>                                                          65536);
> >> Shouldn't the above be fixed as well? Has anybody tried to compile on
> >> 32-bit arch this?
> > No, it correct already. In this case, rem is being typecasted to s64
> > and then multiplied with a 32-bit number, 1000000. Thus 64-bit
> > arithmetic is being performed here.
> 
> What Andy means is that this needs to be DIV_S64_ROUND_CLOSEST() to
> work on 32-bit platforms. But it is clearly unrelated to your change and should
> be in its own patch.

Indeed, I didn't test it on 32 bits.
But both changes make sense to me.

Regards,
Mihail

Re: [PATCH] drivers:iio:dac make expression evaluation 64-bit

Posted by Dan Carpenter 4 years, 5 months ago

On Wed, Dec 22, 2021 at 12:20:32AM +0500, Muhammad Usama Anjum wrote:
> Two 32-bit values are being evaluated using 32-bit arithmetic and then
> passed to s64 type. It is wrong. Expression should be evaluated using
> 64-bit arithmetic and then passed.
> 
> Fixes: 8f2b54824b ("drivers:iio:dac: Add AD3552R driver support")
> Signed-off-by: Muhammad Usama Anjum <usama.anjum@collabora.com>
> ---
>  drivers/iio/dac/ad3552r.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/iio/dac/ad3552r.c b/drivers/iio/dac/ad3552r.c
> index 97f13c0b9631..b03d3c7cd4c4 100644
> --- a/drivers/iio/dac/ad3552r.c
> +++ b/drivers/iio/dac/ad3552r.c
> @@ -770,7 +770,7 @@ static void ad3552r_calc_gain_and_offset(struct ad3552r_desc *dac, s32 ch)
>  	dac->ch_data[ch].scale_dec = DIV_ROUND_CLOSEST((s64)rem * 1000000,
>  							65536);
>  
> -	dac->ch_data[ch].offset_int = div_s64_rem(v_min * 65536, span, &rem);
> +	dac->ch_data[ch].offset_int = div_s64_rem(v_min * 65536L, span, &rem);

"v_min" is relatively close to zero on a number line so this can't
overflow.  There is no way that this change affects anything at runtime
(except making the code a tiny tiny bit slower).

And it should be 65536LL for 32 bit systems?

But I just don't see the point of this change.  Presumably it is to make
a static analyzer happy?

regards,
dan carpenter

RE: [PATCH] drivers:iio:dac make expression evaluation 64-bit

Posted by Sa, Nuno 4 years, 5 months ago

> From: Dan Carpenter <dan.carpenter@oracle.com>
> Sent: Wednesday, January 5, 2022 2:39 PM
> To: Muhammad Usama Anjum <usama.anjum@collabora.com>
> Cc: Lars-Peter Clausen <lars@metafoo.de>; Hennerich, Michael
> <Michael.Hennerich@analog.com>; Jonathan Cameron
> <jic23@kernel.org>; Chindris, Mihail <Mihail.Chindris@analog.com>;
> open list:IIO SUBSYSTEM AND DRIVERS <linux-iio@vger.kernel.org>;
> open list <linux-kernel@vger.kernel.org>; kernel@collabora.com;
> kernel-janitors@vger.kernel.org
> Subject: Re: [PATCH] drivers:iio:dac make expression evaluation 64-bit
> 
> [External]
> 
> On Wed, Dec 22, 2021 at 12:20:32AM +0500, Muhammad Usama Anjum
> wrote:
> > Two 32-bit values are being evaluated using 32-bit arithmetic and
> then
> > passed to s64 type. It is wrong. Expression should be evaluated using
> > 64-bit arithmetic and then passed.
> >
> > Fixes: 8f2b54824b ("drivers:iio:dac: Add AD3552R driver support")
> > Signed-off-by: Muhammad Usama Anjum
> <usama.anjum@collabora.com>
> > ---
> >  drivers/iio/dac/ad3552r.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/iio/dac/ad3552r.c b/drivers/iio/dac/ad3552r.c
> > index 97f13c0b9631..b03d3c7cd4c4 100644
> > --- a/drivers/iio/dac/ad3552r.c
> > +++ b/drivers/iio/dac/ad3552r.c
> > @@ -770,7 +770,7 @@ static void
> ad3552r_calc_gain_and_offset(struct ad3552r_desc *dac, s32 ch)
> >  	dac->ch_data[ch].scale_dec = DIV_ROUND_CLOSEST((s64)rem
> * 1000000,
> >  							65536);
> >
> > -	dac->ch_data[ch].offset_int = div_s64_rem(v_min * 65536,
> span, &rem);
> > +	dac->ch_data[ch].offset_int = div_s64_rem(v_min * 65536L,
> span, &rem);
> 
> "v_min" is relatively close to zero on a number line so this can't
> overflow.  There is no way that this change affects anything at runtime
> (except making the code a tiny tiny bit slower).
> 
> And it should be 65536LL for 32 bit systems?
>

If I'm not missing nothing obvious, 65536LL is the right thing to do...
I did not really checked, but if v_min * 65536 can never overflow, 
then yeah, this is not really "fixing" nothing.

- Nuno Sá