Skip to contentSkip to conversation

iio: chemical: scd30: Cleanup initializations in scd30_float_to_fp()

PATCH Superseded Maxwell Doose 4 participants 7 messages May 8 – May 11

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 iio: chemical: scd30: Cleanup initializations in scd30_float_to_fp()

The current variable declaration and initializations are barely readable and use comma separations across multiple lines. Refactor the initializations so that mantissa and exp have separate declarations and sign gets initialized later.

---
ps:
Hi Jonathan, I noticed a potential divide-by-zero bug on line 241 in scd30_read_raw(), where the value of tmp is dictated by hardware. If the scd30_command_read() call on line 236 assigns 0 to tmp, then when we run:
*val2 = 1000000000 / tmp;
we'll get a divide-by-zero. Will send a patch for this later.

best regards,
max

 drivers/iio/chemical/scd30_core.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
drivers/iio/chemical/scd30_core.c +9 -4
index a665fcb78806..be8c055be184 100644
@@ -89,10 +89,15 @@ static int scd30_reset(struct scd30_state *state)
 /* simplified float to fixed point conversion with a scaling factor of 0.01 */
 static int scd30_float_to_fp(int float32)
 {
-	int fraction, shift,
-	    mantissa = float32 & GENMASK(22, 0),
-	    sign = (float32 & BIT(31)) ? -1 : 1,
-	    exp = (float32 & ~BIT(31)) >> 23;
+	int fraction, shift, sign;
+	int mantissa = float32 & GENMASK(22, 0);
+	int exp = (float32 & ~BIT(31)) >> 23;
+
+	/* Determine sign of received float based on IEEE 754 standard */
+	if (float32 & BIT(31))
+		sign = -1;
+	else
+		sign = 1;
 
 	/* special case 0 */
 	if (!exp && !mantissa)

--
2.54.0

#1 lore raw
Quoted 33 lines · Maxwell Doose

The current variable declaration and initializations are barely readable and use comma separations across multiple lines. Refactor the initializations so that mantissa and exp have separate declarations and sign gets initialized later.

---
ps:
Hi Jonathan, I noticed a potential divide-by-zero bug on line 241 in scd30_read_raw(), where the value of tmp is dictated by hardware. If the scd30_command_read() call on line 236 assigns 0 to tmp, then when we run:
*val2 = 1000000000 / tmp;
we'll get a divide-by-zero. Will send a patch for this later.

best regards,
max

drivers/iio/chemical/scd30_core.c | 13 +++++++++----

1 file changed, 9 insertions(+), 4 deletions(-)
drivers/iio/chemical/scd30_core.c +0 -4
index a665fcb78806..be8c055be184 100644
@@ -89,10 +89,15 @@ static int scd30_reset(struct scd30_state *state)
 /* simplified float to fixed point conversion with a scaling factor of 0.01 */
 static int scd30_float_to_fp(int float32)
 {
-	int fraction, shift,
-	    mantissa = float32 & GENMASK(22, 0),
-	    sign = (float32 & BIT(31)) ? -1 : 1,
-	    exp = (float32 & ~BIT(31)) >> 23;

drivers/iio/chemical/scd30_core.c static int scd30_reset(struct scd30_state *state) · line 101

Quoted 10 lines
drivers/iio/chemical/scd30_core.c +9 -0
+	int fraction, shift, sign;
+	int mantissa = float32 & GENMASK(22, 0);
+	int exp = (float32 & ~BIT(31)) >> 23;
+
+	/* Determine sign of received float based on IEEE 754 standard */
 
+	if (float32 & BIT(31))
+		sign = -1;
+	else
+		sign = 1;

Hi,

Thank you for the refactor change.

The previous version was more compact and readable. Splitting the variable declarations and expanding the sign assignment into an if/else block does not improve clarity significantly.

You can keep it simpler like:

sign = (float32 & BIT(31)) ? -1 : 1;

The IEEE 754 representation is already implicit from the bit manipulation, so additional comment is probably unnecessary.

Thanks, Sanjay

Quoted 6 lines
drivers/iio/chemical/scd30_core.c +0 -1
 	/* special case 0 */
 	if (!exp && !mantissa)
--

2.54.0

#2 lore raw
Quoted 5 lines · Sanjay Chitroda

Thank you for the refactor change.

The previous version was more compact and readable. Splitting the variable declarations and expanding the sign assignment into an if/else block does not improve clarity significantly.

Are you sure? I find it to be far more readable than the comma mess that there was before, plus not a huge fan of the ternary operator in such expressions. One of the goals I have as the new maintainer [1] of the scd30 driver is to get rid of "clever" code and replace it with more explicit and readable code. It's going to get optimized by the compiler anyways. As a wise man once said,

“Programs must be written for people to read, and only incidentally for machines to execute.”

Plus if we need to drop a printk() or whatever in there to see the variable assignment we can just add {} and add the printk(), no messing with the ternary required.

Quoted 5 lines

You can keep it simpler like:

sign = (float32 & BIT(31)) ? -1 : 1;

See above.

Quoted 3 lines

The IEEE 754 representation is already implicit from the bit manipulation, so additional comment is probably unnecessary.

I'll see if Jonathan will want to remove that during merge or if I need to send a v2 (likely not).

best regards,
max

[1] ("MAINTAINERS: Add myself as SCD30 maintainer"):
https://git.kernel.org/pub/scm/linux/kernel/git/jic23/iio.git/commit/?h=togreg

Quoted 7 lines

Thanks, Sanjay

Quoted 6 lines
      /* special case 0 */
      if (!exp && !mantissa)

--
2.54.0

#3 lore raw
Quoted 16 lines · Maxwell Doose
Quoted 6 lines · Sanjay Chitroda

Thank you for the refactor change.

The previous version was more compact and readable. Splitting the variable declarations and expanding the sign assignment into an if/else block does not improve clarity significantly.

Are you sure? I find it to be far more readable than the comma mess that there was before, plus not a huge fan of the ternary operator in such expressions. One of the goals I have as the new maintainer [1] of the scd30 driver is to get rid of "clever" code and replace it with more explicit and readable code. It's going to get optimized by the compiler anyways. As a wise man once said,

Quoted 6 lines

“Programs must be written for people to read, and only incidentally for machines to execute.”

Plus if we need to drop a printk() or whatever in there to see the variable assignment we can just add {} and add the printk(), no messing with the ternary required.

Sanjay, the comma operator is discouraged in the kernel. For this part I'm fully on the Maxwell's side. As for if-else versus ternary it depends on the maintainer, but in general if-else is considered better. (For instance, I'm judging on case-by-case basis, and in some cases ternary makes a lot of sense.) So, in this case I have no strong opinion about sign assignment.

Quoted 8 lines
Quoted 3 lines

You can keep it simpler like:

sign = (float32 & BIT(31)) ? -1 : 1;

See above.

Quoted 2 lines

The IEEE 754 representation is already implicit from the bit manipulation, so additional comment is probably unnecessary.

Since we don't do floats in the kernel (all arithmetics is integer-based) some might find the comment useful. But here as well, no strong opinion.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
Reviewed-by on v1 1/1
#4 lore raw
In reply to Maxwell Doose
Quoted 4 lines · Maxwell Doose

The current variable declaration and initializations are barely readable and use comma separations across multiple lines. Refactor the initializations so that mantissa and exp have separate declarations and sign gets initialized later.

The change LGTM,

...

Quoted 7 lines

ps:
Hi Jonathan, I noticed a potential divide-by-zero bug on line 241 in scd30_read_raw(), where the value of tmp is dictated by hardware. If the scd30_command_read() call on line 236 assigns 0 to tmp, then when we run:
*val2 = 1000000000 / tmp;
we'll get a divide-by-zero. Will send a patch for this later.

Have you checked the Clang and GCC compiled code in that cases? Do they put anything interesting in the binary? (I.o.w. can they prove the division-by-zero at compile-time?)

See the (somehow) related case in the 7e5c0f97c66a ("iio: adc: nxp-sar-adc:
Avoid division by zero"), which is in Linux Next only (for now).

Signature · 2 lines
With Best Regards,
Andy Shevchenko
maintainer
#5 lore raw
In reply to Maxwell Doose
Quoted 6 lines · On Fri, 8 May 2026 17:55:00 -0500 Maxwell Doose <[email protected]> wrote:

The current variable declaration and initializations are barely readable and use comma separations across multiple lines. Refactor the initializations so that mantissa and exp have separate declarations and sign gets initialized later.

This is good but we can take this opportunity to make it more idiomatic - so more 'standard' for new kernel code.

Quoted 8 lines

---
ps:
Hi Jonathan, I noticed a potential divide-by-zero bug on line 241 in scd30_read_raw(), where the value of tmp is dictated by hardware. If the scd30_command_read() call on line 236 assigns 0 to tmp, then when we run:
*val2 = 1000000000 / tmp;
we'll get a divide-by-zero. Will send a patch for this later.

Good to know but not sure why you'd put that comment in this patch. Also as a fix, it should come first - ah so maybe you were indicating it might cause a merge issue?

Quoted 12 lines

best regards,
max

 drivers/iio/chemical/scd30_core.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
drivers/iio/chemical/scd30_core.c +0 -0
index a665fcb78806..be8c055be184 100644

drivers/iio/chemical/scd30_core.c static int scd30_reset(struct scd30_state *state) · line 93

Quoted 9 lines
drivers/iio/chemical/scd30_core.c +2 -4
@@ -89,10 +89,15 @@ static int scd30_reset(struct scd30_state *state)
 /* simplified float to fixed point conversion with a scaling factor of 0.01 */
 static int scd30_float_to_fp(int float32)
 {
-	int fraction, shift,
-	    mantissa = float32 & GENMASK(22, 0),
-	    sign = (float32 & BIT(31)) ? -1 : 1,
-	    exp = (float32 & ~BIT(31)) >> 23;
+	int fraction, shift, sign;
+	int mantissa = float32 & GENMASK(22, 0);
	int mantissa = FIELD_GET(GENMASK(22, 0), float32);

though maybe worth some field defines as well.
//up by defines at top of file.
#define SCD30_FLOAT_MANTISSA_MSK GENMASK(22, 0)

	int mantissa = FIELD_GET(SCD30_FLOAT_MANTISSA_MSK, float32);

drivers/iio/chemical/scd30_core.c static int scd30_reset(struct scd30_state *state) · line 94

Quoted 1 line
drivers/iio/chemical/scd30_core.c +1 -0
+	int exp = (float32 & ~BIT(31)) >> 23;

Again with a field define this can become something like:

//up by defines at top of file - and check carefully I got this right! #define SCD30_FLOAT_EXPONENT_MSK GENMASK(30, 23)

	int exp = FIELD_GET(SCD30_FLOAT_EXPONENT_MSK, float32);

drivers/iio/chemical/scd30_core.c static int scd30_reset(struct scd30_state *state) · line 97

Quoted 3 lines
drivers/iio/chemical/scd30_core.c +3 -0
+
+	/* Determine sign of received float based on IEEE 754 standard */
+	if (float32 & BIT(31))
	if (FIELD_GET(SCD30_FLOAT_SIGN, float32))
Quoted 8 lines
drivers/iio/chemical/scd30_core.c +3 -1
+		sign = -1;
+	else
+		sign = 1;
 
 	/* special case 0 */
 	if (!exp && !mantissa)
--

2.54.0

#6 lore raw
Quoted 9 lines · Jonathan Cameron
Quoted 5 lines · On Fri, 8 May 2026 17:55:00 -0500 Maxwell Doose <[email protected]> wrote:

The current variable declaration and initializations are barely readable and use comma separations across multiple lines. Refactor the initializations so that mantissa and exp have separate declarations and sign gets initialized later.

Quoted 5 lines
Quoted 1 line

This is good but we can take this opportunity to make it more idiomatic - so more 'standard' for new kernel code.

Yea that all sounds good. TBH I'd like to get some of the kernel dev beginners to check out this driver since in some of my earlier emails I said it was messy (+ I'd like to get it cleaned up) and there are probably a metric ton of low hanging fruit.

Quoted 9 lines
Quoted 9 lines

---
ps:
Hi Jonathan, I noticed a potential divide-by-zero bug on line 241 in scd30_read_raw(), where the value of tmp is dictated by hardware. If the scd30_command_read() call on line 236 assigns 0 to tmp, then when we run:
*val2 = 1000000000 / tmp;
we'll get a divide-by-zero. Will send a patch for this later.

Quoted 4 lines

Good to know but not sure why you'd put that comment in this patch. Also as a fix, it should come first - ah so maybe you were indicating it might cause a merge issue?

Well now that you told me that the div-by-zero in the write path was already fixed my v2 would cause a conflict, shouldn't be a problem if I revert it and add some of the other fixes into it.

Quoted 38 lines
Quoted 13 lines

best regards,
max

 drivers/iio/chemical/scd30_core.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
drivers/iio/chemical/scd30_core.c +0 -0
index a665fcb78806..be8c055be184 100644

drivers/iio/chemical/scd30_core.c static int scd30_reset(struct scd30_state *state) · line 93

Quoted 9 lines
drivers/iio/chemical/scd30_core.c +2 -4
@@ -89,10 +89,15 @@ static int scd30_reset(struct scd30_state *state)
 /* simplified float to fixed point conversion with a scaling factor of 0.01 */
 static int scd30_float_to_fp(int float32)
 {
-     int fraction, shift,
-         mantissa = float32 & GENMASK(22, 0),
-         sign = (float32 & BIT(31)) ? -1 : 1,
-         exp = (float32 & ~BIT(31)) >> 23;
+     int fraction, shift, sign;
+     int mantissa = float32 & GENMASK(22, 0);
        int mantissa = FIELD_GET(GENMASK(22, 0), float32);

though maybe worth some field defines as well.
//up by defines at top of file.
#define SCD30_FLOAT_MANTISSA_MSK GENMASK(22, 0)

        int mantissa = FIELD_GET(SCD30_FLOAT_MANTISSA_MSK, float32);

drivers/iio/chemical/scd30_core.c static int scd30_reset(struct scd30_state *state) · line 94

Quoted 1 line
drivers/iio/chemical/scd30_core.c +1 -0
+     int exp = (float32 & ~BIT(31)) >> 23;

Again with a field define this can become something like:

//up by defines at top of file - and check carefully I got this right! #define SCD30_FLOAT_EXPONENT_MSK GENMASK(30, 23)

Quoted 17 lines
        int exp = FIELD_GET(SCD30_FLOAT_EXPONENT_MSK, float32);

drivers/iio/chemical/scd30_core.c static int scd30_reset(struct scd30_state *state) · line 97

Quoted 3 lines
drivers/iio/chemical/scd30_core.c +3 -0
+
+     /* Determine sign of received float based on IEEE 754 standard */
+     if (float32 & BIT(31))
        if (FIELD_GET(SCD30_FLOAT_SIGN, float32))
Quoted 8 lines
drivers/iio/chemical/scd30_core.c +3 -1
+             sign = -1;
+     else
+             sign = 1;
 
      /* special case 0 */
      if (!exp && !mantissa)
--

2.54.0

I'll add those macros for the v2. Also will probably add a comment saying something along the lines of:

/* IEEE 754 floating point arithmetic macros */

best regards,
max