Re: [PATCH 3/9] bitmap: Use FIELD_PREP() in expansion of FIELD_PREP_WM16()

On Wednesday, 10 December 2025 21:59:15 Central European Standard Time David Laight wrote:

On Wed, 10 Dec 2025 20:18:30 +0100
Nicolas Frattaroli [off-list ref] wrote:

quoted

On Tuesday, 9 December 2025 11:03:07 Central European Standard Time david.laight.linux@gmail.com wrote:

quoted

From: David Laight <redacted>

Instead of directly expanding __BF_FIELD_CHECK() (which really ought
not be used outside bitfield) and open-coding the generation of the
masked value, just call FIELD_PREP() and add an extra check for
the mask being at most 16 bits.

Signed-off-by: David Laight <redacted>
---
 include/linux/hw_bitfield.h | 17 ++++++++---------
 1 file changed, 8 insertions(+), 9 deletions(-)

diff --git a/include/linux/hw_bitfield.h b/include/linux/hw_bitfield.h
index df202e167ce4..d7f21b60449b 100644
--- a/include/linux/hw_bitfield.h
+++ b/include/linux/hw_bitfield.h

@@ -23,15 +23,14 @@
  * register, a bit in the lower half is only updated if the corresponding bit
  * in the upper half is high.
  */
-#define FIELD_PREP_WM16(_mask, _val)					     \
-	({								     \
-		typeof(_val) __val = _val;				     \
-		typeof(_mask) __mask = _mask;				     \
-		__BF_FIELD_CHECK(__mask, ((u16)0U), __val,		     \
-				 "HWORD_UPDATE: ");			     \
-		(((typeof(__mask))(__val) << __bf_shf(__mask)) & (__mask)) | \
-		((__mask) << 16);					     \
-	})
+#define FIELD_PREP_WM16(mask, val)				\
+({								\
+	__auto_type _mask = mask;				\
+	u32 _val = FIELD_PREP(_mask, val);			\
+	BUILD_BUG_ON_MSG(_mask > 0xffffu,			\
+			 "FIELD_PREP_WM16: mask too large");	\
+	_val | (_mask << 16);					\
+})
 
 /**
  * FIELD_PREP_WM16_CONST() - prepare a constant bitfield element with a mask in
  

This breaks the build for at least one driver that uses
FIELD_PREP_WM16, namely phy-rockchip-emmc.c:

Not in my allmodconfig build.
... 

quoted

pcie-dw-rockchip.c is similarly broken by this change, except
without the superfluous wrapper:

That one did get built.

I build with clang 21.1.6 for arm64, in case that's any help.
I don't see how pcie-dw-rockchip.c built for you if FIELD_PREP
and FIELD_PREP_WM16 have conflicting symbol names?

The problem is that FIELD_PREP_WM16() needs to use different 'local'
variables than FIELD_PREP().
The 'proper' fix is to use unique names (as min() and max() do), but that
makes the whole thing unreadable and is best avoided unless nesting is
likely.
In this case s/mask/wm16_mask/ and s/val/wm16_val/ might be best.

	David