On 10/10/2018 19:13, Johannes Berg wrote:
On Wed, 2018-10-10 at 10:33 -0700, Joe Perches wrote:
Specifically it doesn't like the __BF_FIELD_CHECK() in FIELD_PREP().
Any ideas on compiler trickery we could do with the FIELD_PREP()
definition to avoid this issue (i.e. enforce the check but only use the
constant value)?
thanks guys
Perhaps __bf_shf should not use __builtin_ffsll.
__bf_shf() is a constant expression, and is fine in this context.
The problem is the use of the compound statement here:
static int x[2] = {
({ (void)(0); 1; }),
0,
}
similarly fails to compile.
I've recently run into a similar situation, namely in
include/net/netlink.h, and the applicable way to solve it here would be
something like this:
diff --git a/include/linux/bitfield.h b/include/linux/bitfield.h
index 3f1ef4450a7c..0680d641923f 100644
--- a/include/linux/bitfield.h
+++ b/include/linux/bitfield.h
@@ -49,19 +49,16 @@
#define __bf_shf(x) (__builtin_ffsll(x) - 1)
+#define BUILD_BUG_ON_RET_ZERO(cond) (sizeof(char[1 - 2*!!(cond)]) - 1)
+#define BUILD_BUG_ON_NOT_POW2_RET_ZERO(n) BUILD_BUG_ON_RET_ZERO(((n) & ((n) - 1)) != 0)
+
#define __BF_FIELD_CHECK(_mask, _reg, _val, _pfx) \
- ({ \
- BUILD_BUG_ON_MSG(!__builtin_constant_p(_mask), \
- _pfx "mask is not constant"); \
- BUILD_BUG_ON_MSG((_mask) == 0, _pfx "mask is zero"); \
- BUILD_BUG_ON_MSG(__builtin_constant_p(_val) ? \
- ~((_mask) >> __bf_shf(_mask)) & (_val) : 0, \
- _pfx "value too large for the field"); \
- BUILD_BUG_ON_MSG((_mask) > (typeof(_reg))~0ull, \
- _pfx "type of reg too small for mask"); \
- __BUILD_BUG_ON_NOT_POWER_OF_2((_mask) + \
- (1ULL << __bf_shf(_mask))); \
- })
+ BUILD_BUG_ON_RET_ZERO(!__builtin_constant_p(_mask)) + \
+ BUILD_BUG_ON_RET_ZERO((_mask) == 0) + \
+ BUILD_BUG_ON_RET_ZERO(__builtin_constant_p(_val) ? \
+ ~((_mask) >> __bf_shf(_mask)) & (_val) : 0) + \
+ BUILD_BUG_ON_RET_ZERO((_mask) > (typeof(_reg))~0ull) + \
+ BUILD_BUG_ON_NOT_POW2_RET_ZERO((_mask) + (1ULL << __bf_shf(_mask)))
/**
* FIELD_FIT() - check if value fits in the field
@@ -85,10 +82,8 @@
* be combined with other fields of the bitfield using logical OR.
*/
#define FIELD_PREP(_mask, _val) \
- ({ \
- __BF_FIELD_CHECK(_mask, 0ULL, _val, "FIELD_PREP: "); \
- ((typeof(_mask))(_val) << __bf_shf(_mask)) & (_mask); \
- })
+ (__BF_FIELD_CHECK(_mask, 0ULL, _val, "FIELD_PREP: ") + \
+ (((typeof(_mask))(_val) << __bf_shf(_mask)) & (_mask)))
/**
* FIELD_GET() - extract a bitfield element
Note that this is an incomplete patch - everything but FIELD_PREP will
not compile with this.
Also, BUILD_BUG_ON_RET_ZERO and BUILD_BUG_ON_NOT_POW2_RET_ZERO should
probably have better names, or perhaps do the positive way that I did in
__NLA_ENSURE, e.g. CONST_ASSERT()/CONST_ASSERT_IS_POWER_OF_2()? I guess
they should go to build_bug.h as well...
Seems reasonable. However I did try this and was getting compiler
warnings about VLA, from a non-constant being fed into
BUILD_BUG_ON_RET_ZERO(), related to sizeof char[]:
drivers/iio/adc/meson_saradc.c:375:2: warning: ISO C90 forbids variable
length array [-Wvla]
regval = FIELD_PREP(MESON_SAR_ADC_CHAN_LIST_ENTRY_MASK(0),
Surely __NLA_ENSURE is getting a similar issue as it uses a similar
principle, no? I see that this is in -next now, but could not this macro
or derivatives being referenced.
Much appreciated,
John
johannes
.
