[v7,2/5] bits: introduce fixed-type BIT_U*()

Commit Message

Vincent Mailhol March 22, 2025, 9:23 a.m. UTC

From: Lucas De Marchi <[e-mail]>

Implement fixed-type BIT_U*() to help drivers add stricter checks,
like it was done for GENMASK_U*().

Signed-off-by: Lucas De Marchi <[e-mail]>
Acked-by: Jani Nikula <[e-mail]>
Co-developed-by: Vincent Mailhol <[e-mail]>
Signed-off-by: Vincent Mailhol <[e-mail]>
---
Changelog:

  v6 -> v7:
  v5 -> v6:

    - No changes.

  v4 -> v5:

    - Rename GENMASK_t() to GENMASK_TYPE().

    - Use tab indentations instead of single space to separate the
      macro name from its body.

    - Add a global comment at the beginning of the file to explain why
      GENMASK_U*() and BIT_U*() are not available in asm.

    - Add a new BIT_TYPE() helper function, similar to GENMASK_TYPE().

    - Remove the unsigned int cast for the U8 and U16 variants. Move
      the cast to BIT_TYPE().

    - Rename the argument from BIT_U*(b) to BIT_U*(nr) for consistency
      with vdso/bits.h.

  v3 -> v4:

    - Use const_true() to simplify BIT_INPUT_CHECK().

    - Make BIT_U8() and BIT_U16() return an unsigned int instead of a
      u8 and u16. Because of the integer promotion rules in C, an u8
      or an u16 would become a signed integer as soon as these are
      used in any expression. By casting these to unsigned ints, at
      least the signedness is kept.

    - Put the cast next to the BIT() macro.

    - In BIT_U64(): use BIT_ULL() instead of BIT().
---
 include/linux/bits.h | 20 +++++++++++++++++++-
 1 file changed, 19 insertions(+), 1 deletion(-)

Comments

Andy Shevchenko March 24, 2025, 1:41 p.m. UTC | #1

On Sat, Mar 22, 2025 at 06:23:13PM +0900, Vincent Mailhol via B4 Relay wrote:
> From: Lucas De Marchi <[e-mail]>
> 
> Implement fixed-type BIT_U*() to help drivers add stricter checks,
> like it was done for GENMASK_U*().

...

> +/*
> + * Fixed-type variants of BIT(), with additional checks like GENMASK_TYPE(). The
> + * following examples generate compiler warnings due to shift-count-overflow:

"...due to -Wshift-count-overflow:" ?

Same idea — if you need a new version, since it's just a nit-pick.

> + *
> + * - BIT_U8(8)
> + * - BIT_U32(-1)
> + * - BIT_U32(40)
> + */

Vincent Mailhol March 24, 2025, 2:16 p.m. UTC | #2

On 24/03/2025 at 22:41, Andy Shevchenko wrote:
> On Sat, Mar 22, 2025 at 06:23:13PM +0900, Vincent Mailhol via B4 Relay wrote:
>> From: Lucas De Marchi <[e-mail]>
>>
>> Implement fixed-type BIT_U*() to help drivers add stricter checks,
>> like it was done for GENMASK_U*().
> 
> ...
> 
>> +/*
>> + * Fixed-type variants of BIT(), with additional checks like GENMASK_TYPE(). The
>> + * following examples generate compiler warnings due to shift-count-overflow:
> 
> "...due to -Wshift-count-overflow:" ?
> 
> Same idea — if you need a new version, since it's just a nit-pick.

If you want. I staged this change locally, so if there is a v8, it will
be addressed. I applied the same to the previous patch which also
mentioned shift-count-overflow without the -W prefix.

But honestly, I am not convinced of the added value. This is from Lucas
original patch one year ago, and no one was bothered by this. IMHO, when
writing:

  (...) generate compiler warnings due to shift-count-overflow:

I do not see where the ambiguity is. The sentence clearly say that this
is a compiler warning, so with or without the -W prefix, the sentence is
equally understandable.


Yours sincerely,
Vincent Mailhol

Andy Shevchenko March 24, 2025, 2:32 p.m. UTC | #3

On Mon, Mar 24, 2025 at 11:16:30PM +0900, Vincent Mailhol wrote:
> On 24/03/2025 at 22:41, Andy Shevchenko wrote:
> > On Sat, Mar 22, 2025 at 06:23:13PM +0900, Vincent Mailhol via B4 Relay wrote:


...

> >> +/*
> >> + * Fixed-type variants of BIT(), with additional checks like GENMASK_TYPE(). The
> >> + * following examples generate compiler warnings due to shift-count-overflow:
> > 
> > "...due to -Wshift-count-overflow:" ?
> > 
> > Same idea — if you need a new version, since it's just a nit-pick.
> 
> If you want. I staged this change locally, so if there is a v8, it will
> be addressed. I applied the same to the previous patch which also
> mentioned shift-count-overflow without the -W prefix.
> 
> But honestly, I am not convinced of the added value. This is from Lucas
> original patch one year ago, and no one was bothered by this. IMHO, when
> writing:
> 
>   (...) generate compiler warnings due to shift-count-overflow:
> 
> I do not see where the ambiguity is. The sentence clearly say that this
> is a compiler warning, so with or without the -W prefix, the sentence is
> equally understandable.

As I marked, it's a nit-pick, but from my point of view the added value
is immediate: The reader can be sure that we are talking about a compiler
warning and not something else (C standard? some special term?). So it adds
more context and makes it clearer.

Patch

diff --git a/include/linux/bits.h b/include/linux/bits.h
index beb3ee2f1bc74a9346dd72eb06c722a9bc536051..6a942ea9ab380d3bd0e521916caa1d59db8031c0 100644
--- a/include/linux/bits.h
+++ b/include/linux/bits.h
@@ -24,7 +24,7 @@ 
 /*
  * Missing asm support
  *
- * GENMASK_U*() depend on BITS_PER_TYPE() which relies on sizeof(),
+ * GENMASK_U*() and BIT_U*() depend on BITS_PER_TYPE() which relies on sizeof(),
  * something not available in asm. Nevertheless, fixed width integers is a C
  * concept. Assembly code can rely on the long and long long versions instead.
  */
@@ -55,6 +55,24 @@ 
 #define GENMASK_U32(h, l)	GENMASK_TYPE(u32, h, l)
 #define GENMASK_U64(h, l)	GENMASK_TYPE(u64, h, l)
 
+/*
+ * Fixed-type variants of BIT(), with additional checks like GENMASK_TYPE(). The
+ * following examples generate compiler warnings due to shift-count-overflow:
+ *
+ * - BIT_U8(8)
+ * - BIT_U32(-1)
+ * - BIT_U32(40)
+ */
+#define BIT_INPUT_CHECK(type, nr) \
+	BUILD_BUG_ON_ZERO(const_true((nr) >= BITS_PER_TYPE(type)))
+
+#define BIT_TYPE(type, nr) ((type)(BIT_INPUT_CHECK(type, nr) + BIT_ULL(nr)))
+
+#define BIT_U8(nr)	BIT_TYPE(u8, nr)
+#define BIT_U16(nr)	BIT_TYPE(u16, nr)
+#define BIT_U32(nr)	BIT_TYPE(u32, nr)
+#define BIT_U64(nr)	BIT_TYPE(u64, nr)
+
 #else /* defined(__ASSEMBLY__) */
 
 /*