[Intel-gfx] [Intel-xe] [PATCH 2/3] linux/bits.h: Add fixed-width GENMASK and BIT macros

[Lucas De Marchi]

On Tue, May 09, 2023 at 11:00:36AM -0300, Gustavo Sousa wrote:
>Quoting Lucas De Marchi (2023-05-09 02:14:02)
>>Add GENMASK_U32(), GENMASK_U16() and GENMASK_U8()  macros to create
>>masks for fixed-width types and also the corresponding BIT_U32(),
>>BIT_U16() and BIT_U8().
>>
>>All of those depend on a new "U" suffix added to the integer constant.
>>Due to naming clashes it's better to call the macro U32. Since C doesn't
>>have a proper suffix for short and char types, the U16 and U18 variants
>>just use U32 with one additional check in the BIT_* macros to make
>>sure the compiler gives an error when the those types overflow.
>>The BIT_U16() and BIT_U8() need the help of GENMASK_INPUT_CHECK(),
>>as otherwise they would allow an invalid bit to be passed. Hence
>>implement them in include/linux/bits.h rather than together with
>>the other BIT* variants.
>>
>>The following test file is is used to test this:
>>
>>        $ cat mask.c
>>        #include <linux/types.h>
>>        #include <linux/bits.h>
>>
>>        static const u32 a = GENMASK_U32(31, 0);
>>        static const u16 b = GENMASK_U16(15, 0);
>>        static const u8 c = GENMASK_U8(7, 0);
>>        static const u32 x = BIT_U32(31);
>>        static const u16 y = BIT_U16(15);
>>        static const u8 z = BIT_U8(7);
>>
>>        #if FAIL
>>        static const u32 a2 = GENMASK_U32(32, 0);
>>        static const u16 b2 = GENMASK_U16(16, 0);
>>        static const u8 c2 = GENMASK_U8(8, 0);
>>        static const u32 x2 = BIT_U32(32);
>>        static const u16 y2 = BIT_U16(16);
>>        static const u8 z2 = BIT_U8(8);
>>        #endif
>>
>>Signed-off-by: Lucas De Marchi <lucas.demarchi at intel.com>
>>---
>> include/linux/bits.h       | 22 ++++++++++++++++++++++
>> include/uapi/linux/const.h |  2 ++
>> include/vdso/const.h       |  1 +
>> 3 files changed, 25 insertions(+)
>>
>>diff --git a/include/linux/bits.h b/include/linux/bits.h
>>index 7c0cf5031abe..ff4786c99b8c 100644
>>--- a/include/linux/bits.h
>>+++ b/include/linux/bits.h
>>@@ -42,4 +42,26 @@
>> #define GENMASK_ULL(h, l) \
>>        (GENMASK_INPUT_CHECK(h, l) + __GENMASK_ULL(h, l))
>>
>>+#define __GENMASK_U32(h, l) \
>>+  (((~U32(0)) - (U32(1) << (l)) + 1) & \
>>+   (~U32(0) >> (32 - 1 - (h))))
>>+#define GENMASK_U32(h, l) \
>>+  (GENMASK_INPUT_CHECK(h, l) + __GENMASK_U32(h, l))
>>+
>>+#define __GENMASK_U16(h, l) \
>>+  ((U32(0xffff) - (U32(1) << (l)) + 1) & \
>>+   (U32(0xffff) >> (16 - 1 - (h))))
>>+#define GENMASK_U16(h, l) \
>>+  (GENMASK_INPUT_CHECK(h, l) + __GENMASK_U16(h, l))
>>+
>>+#define __GENMASK_U8(h, l) \
>>+  (((U32(0xff)) - (U32(1) << (l)) + 1) & \
>>+   (U32(0xff) >> (8 - 1 - (h))))
>>+#define GENMASK_U8(h, l) \
>>+  (GENMASK_INPUT_CHECK(h, l) + __GENMASK_U8(h, l))
>
>I wonder if we should use BIT_U* variants in the above to ensure the values are
>valid. If so, we get a nice boundary check and we also can use a single
>definition for the mask generation:
>
>  #define __GENMASK_U32(h, l) \
>          (((~U32(0)) - (U32(1) << (l)) + 1) & \
>           (~U32(0) >> (32 - 1 - (h))))

the boundary for h and l are already covered here because (32 - 1 - (h))
would lead to a negative value if h >= 32. Similar reason for l

Doing ~U32(0) didn't work for me as it wouldn't catch the invalid values
due to expanding to U32_MAX


>  #define GENMASK_U32(h, l) \
>          (GENMASK_INPUT_CHECK(h, l) + __GENMASK_U32(BIT_U32(h), BIT_U32(l)))

							^^^^
that doesn't really work as BIT_U32(h) would expand here,
creating the equivalent of

	~U32(0) >> (32 - 1 - (BIT_U32(h))),

which is not what we want

>  #define GENMASK_U16(h, l) \
>          (GENMASK_INPUT_CHECK(h, l) + __GENMASK_U32(BIT_U16(h), BIT_U16(l)))
>  #define GENMASK_U8(h, l) \
>          (GENMASK_INPUT_CHECK(h, l) + __GENMASK_U32(BIT_U8(h), BIT_U8(l)))
>
>>+
>>+#define BIT_U32(nr)       _BITU32(nr)
>>+#define BIT_U16(nr)       (GENMASK_INPUT_CHECK(16 - 1, nr) + (U32(1) << (nr)))
>>+#define BIT_U8(nr)        (GENMASK_INPUT_CHECK(32 - 1, nr) + (U32(1) << (nr)))
>
>Shouldn't this be GENMASK_INPUT_CHECK(8 - 1, nr)?

ugh, good catch. Thanks

I will think if I can come up with something that reuses a single
__GENMASK_U(). Meanwhile I improved my negative tests to cover more
cases.

Lucas De Marchi



>
>--
>Gustavo Sousa
>
>>+
>> #endif /* __LINUX_BITS_H */
>>diff --git a/include/uapi/linux/const.h b/include/uapi/linux/const.h
>>index a429381e7ca5..3a4e152520f4 100644
>>--- a/include/uapi/linux/const.h
>>+++ b/include/uapi/linux/const.h
>>@@ -22,9 +22,11 @@
>> #define _AT(T,X)       ((T)(X))
>> #endif
>>
>>+#define _U32(x)           (_AC(x, U))
>> #define _UL(x)         (_AC(x, UL))
>> #define _ULL(x)                (_AC(x, ULL))
>>
>>+#define _BITU32(x)        (_U32(1) << (x))
>> #define _BITUL(x)      (_UL(1) << (x))
>> #define _BITULL(x)     (_ULL(1) << (x))
>>
>>diff --git a/include/vdso/const.h b/include/vdso/const.h
>>index 94b385ad438d..417384a9795b 100644
>>--- a/include/vdso/const.h
>>+++ b/include/vdso/const.h
>>@@ -4,6 +4,7 @@
>>
>> #include <uapi/linux/const.h>
>>
>>+#define U32(x)            (_U32(x))
>> #define UL(x)          (_UL(x))
>> #define ULL(x)         (_ULL(x))
>>
>>--
>>2.40.1
>>